Emergence of evolutionary driving forces in pattern-forming microbial populations

Abstract

Evolutionary dynamics are controlled by a number of driving forces, such as natural selection, random genetic drift and dispersal. In this perspective article, we aim to emphasize that these forces act at the population level, and that it is a challenge to understand how they emerge from the stochastic and deterministic behaviour of individual cells. Even the most basic steric interactions between neighbouring cells can couple evolutionary outcomes of otherwise unrelated individuals, thereby weakening natural selection and enhancing random genetic drift. Using microbial examples of varying degrees of complexity, we demonstrate how strongly cell–cell interactions influence evolutionary dynamics, especially in pattern-forming systems. As pattern formation itself is subject to evolution, we propose to study the feedback between pattern formation and evolutionary dynamics, which could be key to predicting and potentially steering evolutionary processes. Such an effort requires extending the systems biology approach from the cellular to the population scale.

This article is part of the theme issue ‘Self-organization in cell biology’.

1. Introduction

One of the great challenges of living systems is the orchestration of their building blocks. Many of them employ self-organization by which functional structures emerge solely from interactions among lower-level components [1–5]. The power of self-organization stems from its simplicity—a complex orchestrating machinery is not necessary. This makes self-organization easy to implement and robust to fluctuations. Classic examples can be found from molecular to ecosystem scales, including the organization of growth and division of cells [2,6], the development of multi-cellular entities [7] such as microbial communities [8,9] or the collective motion of bacteria [10].

The principles of pattern formation have been studied intensively in physics and engineering, where mathematical modelling is used to predict the behaviour of large systems of interacting components [11–15]. These approaches have made great strides in explaining and predicting self-organized structures in biology, which have often been verified by experiments [1,2,6,16–20]. Interestingly, even quite complex structures can result from simple behaviours performed by individuals relying on local information only, as epitomized by Turing patterns (figure 1a), fruiting bodies of slime moulds (figure 1b) and surface wrinkles on a biofilm (figure 1c) [21–23].

Figure 1.

Self-organization generates patterns during the development of multi-cellular systems. (a) Dynamical Turing patterns form on lizard skins [21]. (b) Tens of thousands of cells of the slime mould D. discoideum come together to generate a fruiting body [22]. (c) B. subtilis biofilms exhibit a complex pattern of three-dimensional surface wrinkles [23]. (d) Gene segregation in colonies of the bacterium E. coli results in a pattern of rugged sectors [24].

In this perspective article, we explore biological pattern formation in light of Theodosius Dobzhansky's motto, ‘nothing in biology makes sense except in the light of evolution’ [25]. It is natural to suspect evolution to have tinkered with any extant biological pattern, either because certain patterns are selected for their benefit to the pattern-generating organisms or as a side product of selection for the fittest. But one may also ask the reverse question: does pattern formation influence evolutionary dynamics?

For the special case of the evolution of cooperation, a link to pattern formation is well documented [26]. Spatio-temporal structures can promote or disrupt the cohesiveness of groups of cooperators, and thereby influence the invasion probability of defectors. But recent microbial studies, which we survey below, suggest that even the most basic Darwinian principles are susceptible to pattern formation: patterns generated by self-organization determine the proliferation and movement of individual cells in space and time. This in turn generates associations among lineages and environments, which can strongly influence the influx of new mutations, the competition between genotypes and the strength of genetic drift. These observations underscore the fact that the key forces of evolution act at the population level: natural selection and genetic drift characterize the collective behaviour and relationships of many individuals [27]. They cannot easily be intuited from single-cell properties.

We also discuss microbial systems in which one can watch evolutionary dynamics massively changing pattern formation in a few generations. Those cases make evident that self-organization and evolutionary dynamics shape each other and, in general, need to be understood jointly. Revealing this feedback loop could be key to several lines of inquiry: can one predict how evolutionary forces originate from and are modulated by microscopic cell–cell interactions and ensuing self-organization? To what extent are forces of evolution similar among populations of widely differing organisms and in different environments? How can we describe these emergent forces by predictive models on meso-scales? Can we use this knowledge to control harmful evolutionary processes, such as antibiotic-resistance evolution, cancer or epidemic spread? We argue that answering these questions requires extending the systems biology approach from single cells to populations.

2. From cellular stochasticity to macroscopic genetic drift

Population geneticists have long valued the role of chance in evolution: novel mutations can go extinct by chance if their bearers are unlucky and fail to reproduce. Stochastic extinction is, in fact, the typical fate of a mutation even if it confers a slight fitness advantage [28]. Random number fluctuations are therefore considered one of evolution's major driving forces and are conventionally called ‘random genetic drift’. Yet, the system specific determinants of the strength of genetic drift are often elusive.

Classically, random genetic drift is modelled by assuming that offspring numbers exhibit some amount of random variability. This variability is usually assumed to be not correlated among generations—otherwise, it would look heritable and act like natural selection. On this standard view of genetic drift, allele frequencies should fluctuate only weakly in large populations and be primarily controlled by deterministic forces such as natural selection.

However, in pattern-forming systems offspring numbers can become strongly correlated in time and space such that genetic drift can be the dominant force even in very large populations. This can be best appreciated in microbial colonies, where enhanced genetic drift leads to the rapid demixing of initially present genotypes, despite population sizes of up to 10⁹ cells [29] (figure 1d).

The basic reason for enhanced genetic drift is a self-organized bottleneck at the leading edge of the colony [30], and more generally of travelling population waves [31–33]. Pioneer individuals residing in the leading edge have a massive evolutionary advantage over bulk individuals, not only because they have a larger growth rate but, more importantly, because their descendants have a good chance to remain at the expanding frontier where they also enjoy large growth rates (figure 2a–c). This leads to a kind of chain reaction, called ‘gene surfing’, by which pioneers give rise to pioneers of the next generation, and so on [32,35,36]. As a result, mutants can locally fix at the advancing frontier, and give rise to large sectors [29,37]. The effect is so strong that it is a challenge to envision mechanisms that could avoid this local loss of diversity in range expansions [38].

Figure 2.

Interplay of cell–cell interactions and pattern formation impacts evolution on the population scale. (a)–(c) Cell shape influences mechanical interactions on a microscopic level. Rod-like bacteria, such as E. coli (bottom), have a propensity to align into nematic domains, which can increase the lateral dynamics of individual cells at the population front compared to ellipsoidal cells, such as budding yeast (top) [24]. (b) Probability of a lineage to still be at the front as a function of the founder cell's distance to the front. Only cells within one cell diameter of the front have a significant probability to generate long-lasting lineages. (d) The front roughness is linked to the ‘surfing’ probability of a single spontaneous beneficial mutations arising at the front [34]. (e) A positive feedback between position, nutrient concentration and growth rate leads to an amplification of spontaneous front undulations. Light green layer indicates actively dividing cells and orange arrows illustrate resulting local front velocities [34]. (f) On the population scale, the difference in surfing probability is reflected in the number of clonal sectors arising from an initial mixture of strains [24].

The stochastic growth of sectors is strongly influenced by the roughness of the expanding front (figure 2d), which in simulations has been shown to be sensitive to a range of microscopic properties, such as cell shape or anisotropic friction [34]. Indeed, comparing cells of different shapes, for example, ellipsoidal yeast cells to those of rod-like bacteria, shows that rod-like cells can display an increased level of genetic drift (figure 2c) [24].

Front roughness can be amplified further by front instabilities: initial stochastic undulations in the progressing front cause spatial inhomogeneities in the consumption and concentration of nutrients. As a result, cells in a protruding section can profit from higher nutrient concentrations than their counterparts situated in a depression. The ensuing difference in growth rates then amplifies the front undulations, which further amplify growth rate differences. This positive feedback, called fingering instability, generates characteristic dome-like structures at the advancing frontier (figure 2e) [9,39,40]. Importantly, whether a mutant's lineage originates in a protrusion, where it enjoys large growth rates, or in a lagging, growth depressed region is largely random, further enhancing the strength of genetic drift [34].

These and related findings have led to extensive investigations of the patterns of genetic diversity generated in colonies and biofilms, the spatially structured form adopted by many microbial populations in nature [29,37,41–48]. Theoretical studies have explored how the spatio-temporal correlation phenomena emerging in self-organized travelling waves influence neutral evolutionary dynamics [31–33,35,36,38,49–52].

Returning to the fundamental importance of genetic drift, one may now ask how emergent genetic drift competes with selection. When faster growing mutants are successful in the context of a growing microbial colony, they are expected to generate characteristic bulges that gradually take over the expansion front [41,53–55]. However, even beneficial mutants frequently go extinct due to the stochastic wandering of sector boundaries that tends to expel individual lineages from the expanding edge. As a result, enhanced genetic drift severely decreases the probability of an advantageous mutant clone to establish at the front of a macroscopically expanding colony. In experiments with fluorescent mutant-wild-type mixtures, only few initially present mutants are able to survive genetic drift and establish a growing sector bulge even if selection is relatively strong (figure 2f) [24]. The lineages that do manage to establish, however, can rapidly grow to a large size. This is in contrast to an equivalent well-mixed experiment where the number of establishing mutants would be several orders of magnitude higher with each mutant lineage remaining relatively small.

3. Natural selection as an emergent phenomenon

So far, we have discussed patterns formed by phenotypically identical cells. Those ‘neutral’ patterns can enhance random genetic drift, and may shift the balance between genetic drift and natural selection, the other major driving force in evolution. But neutral processes do not influence natural selection per se, which means, for instance, that the long-term deterministic behaviour of a beneficial mutant is not affected by an increase in genetic drift. One may thus wonder whether the character of natural selection is susceptible to pattern formation that acts on phenotypically different cells.

This general question can again be explored using the example of dense microbial communities. We begin with an intuitive argument that suggests that even the simplest, steric, interactions between neighbouring cells can generate spatio-temporal correlations that influence natural selection: when cells divide and grow within a crowded population, they have to exert forces to create the void for their offspring (figure 3a) [57]. These pushing forces, in turn, lead to cell movements and the expansion of the front. Importantly, the motion of any focal cell depends on the forces generated by its mechanically coupled neighbourhood. This is most significant for pioneer cells at the expanding frontier as their motion depends on the collective pushing of the cells behind them. The collective propulsion of pioneer cells represents a kind of mechanically induced ‘cooperation’ that determines which lineages remain growing at the expanding frontier and at what rate [53,58].

Figure 3.

Mechanical cooperation alters natural selection. (a) A budding yeast cell (red) exerts forces on neighbouring cells (arrows) (adapted from [56]). (b) In an expanding yeast colony (advancing edge on top), grown from a faster-growing red strain and a slower-growing yellow strain, collective force generation leads to long-range mechanical interactions resulting in an elongated funnel-shaped sector (see the electronic supplementary material for details). The initial width of the sector was measured after 24 h of colony pre-growth. Coloured arrows illustrate how the observed clonal extinction (blue) is delayed (orange) with respect to extrapolated straight initial boundary trajectories predicted by non-mechanical models (green). Arrow lengths are adjusted to the discernible length of the sector for better illustration (for actual values refer to panel (c)). Grey lines show front position at the indicated times after pre-growth. (c) Quantitative comparison of the observed sector lengths of six independent sectors with those expected in the absence of the mechanical cooperation (see electronic supplementary material for data table and methods). The dashed grey line indicates analytical predictions based on a constant front speed. All other colour labels are as in (b). The sector depicted in panel (b) is indicated with a black triangle.

To demonstrate how this effect can generate an unusual form of selection, we grew a colony from two strains of the non-motile yeast S. cerevisiae where faster-growing ‘wild-type’ cells (red) surround a segment consisting of slower-growing ‘mutant’ cells (yellow) (figure 3b,c; see Materials and Methods in the electronic supplementary material). As the colony expanded, the deleterious mutant initially shrunk in width at an approximately constant rate, congruent with previous non-mechanical models (see electronic supplementary material for a detailed description) [50,53,54]. When the two sector boundaries approach, however, their directions become increasingly correlated (figure 3b). This alignment leads to an inverted funnel-like shape, substantially delaying the clone's extinction compared to the non-mechanical expectation. A similar picture applies to sectors of beneficial mutants, which show a long transient of very slow growth before they reach a large enough size to follow the non-mechanical expectation [54].

We propose that this unexpected behaviour is a direct result of the collective pushing of cells discussed above: their growth-induced mechanical cooperation gradually couples the trajectories of converging lineages, including those at the clone boundaries. The ensuing collective motion of cells effectively conceals the fitness difference of small mutant clones and reduces the strength of selection acting on them (see [59] for a more in-depth study). Broader evolutionary consequences could be (i) an increased genetic load from deleterious mutations, which are weeded out inefficiently [48,53,60,61] and (ii) a lower rate of adaptation from beneficial de novo mutations (since their advantage is initially hampered). Importantly, the underlying mechanical cooperation should be a widespread mechanism as it merely requires growth-induced pushing forces between cells, which easily arises in dense populations, including biofilms, certain tissues and solid tumours.

The induced correlations between lineages depend on the details of the mechanical interaction between cells, which itself varies with cell shape and cell surface properties [62]. For instance, elongated cells, such as rod-like bacteria or ellipsoidal varieties of yeast, have a tendency to align with their neighbours. This can lead to local nematic order inside dense microbial populations (figure 4a, [63]), which can also influence evolutionary outcomes.

Figure 4.

Self-organization due to cell shape impacts evolutionary outcomes. (a) Rod-shaped bacteria align under spatial constraints (adapted from [63], copyright 2008 National Academy of Sciences). (b) Owing to the alignment of their axes of division, rod-shaped cells (blue) burrow underneath a subpopulation of spherical cells (red) in a surface attached in silico colony (top, right; indicated regions are magnified on the left) [64]. The presence of a nutrient gradient in such a stratified colony leads to a coupling of self-organization and evolutionary outcome (bottom, right). (c) Internal pressure in surface attached micro-colonies of V. cholerae causes cells to re-orient from an initial horizontal orientation to one perpendicular to the surface, boosting three-dimensional growth [65].

For example, Smith et al. recently described in a study combining simulations and experiments how cell alignment can lead to stratification in communities constituted of differently shaped cells (figure 4b) [64]. If grown on a surface, elongated cells burrow underneath cells of a more spherical phenotype, lifting them to the upper layers. This cell shape sorting can turn into selection when it is combined with resource gradients. For instance, typical oxygen gradients tend to favour the top layers where the more spherical cells are located. On the other hand, nutrient gradients diffusing into the colony from the attachment surface would favour the bottom layer of elongated cells.

A related nematic ordering effect has recently been observed in micro-colonies of V. cholerae. In this case, growth-induced forces conspire with matrix-mediated cell-to-surface adhesion forces to rotate elongated cells into an out-of-plane orientation as a response to increased local pressure [65]. As a consequence, these biofilms take dome-like rather than flat structure (figure 4c). This two- to three-dimensional transition may thus also lead to a type of spatial cell sorting that can become subject to selection in the presence of resource gradients.

4. Reciprocal interdependence of spatial patterns and evolutionary dynamics

While pattern formation can shape evolutionary forces, the latter can also feed back on processes of self-organization. This can occur in one of two ways. The first is through generic higher fitness mutations, which can change the local community structure via the biophysical interactions discussed in previous sections. For instance, a sector formed by a faster growing mutant lineage in a microbial colony will bulge out and subsequently reshape the front (figure 5a). These changes in front shape can then feed back on future evolutionary processes by altering the establishment probability of a subsequent mutation [34,41].

Figure 5.

Evolution has an impact on formation of spatial patterns. (a) A faster growing lineage (dark) forms a sector that bulges out of the wild-type (yellow) population in a colony of S. cerevisiae, reshaping its front [54]. (b) Evolution of P. aeruginosa on a surface results in ‘hyperswarmers’ with increased mobility and rounder, less dendritic colonies [66]. (c) Selection of B. cenocepacia for surface attachment and subsequent dispersal produces three distinct self-organized colony shapes: (from left to right) studded, ruffled and wrinkly [67]. (d) Selection for rapidly settling yeast leads to the development of multicellular clusters (i). Further adaptation is facilitated through isogenic branches that separate from the main cluster, resulting in evolution of larger cluster sizes (ii) [68].

A second way that evolutionary forces feed back onto pattern formation is through traits that directly affect spatial organization such as motility, adhesion and mechanical interactions between mothers and daughters. The evolution of motility has been shown to affect the spatial structure of swarming P. aeruginosa communities, which can evolve ‘hyperswarmer’ phenotypes with multiple flagella when selected for propagation speed [66]. This transition from wild-type swarmers to hyperswarmers changes the shape of the colony from a branched pattern to a more homogeneous morphology (figure 5b).

Furthermore, Traverse et al. [67] evolved B. cenocepacia by co-selecting for surface adhesion and subsequent redispersal. Their evolution experiments yielded three coexisting phenotypes with distinct modes of self-organized colony morphology: studded, ruffled and wrinkly (figure 5c). Considering the impact of front roughness on evolutionary processes, as discussed in §2, these contrasting morphologies could result in different subsequent evolution within each subpopulation.

Local pattern formation on the length scale of a few tens of cells strongly depends on the mechanical interactions between cells and their direct offsprings, which can also be subject to evolution. An experiment done by Ratcliff et al. in budding yeast studied changes in septation between mother and daughter cells by selecting for rapid sedimentation in liquid media [69]. As a result, the cells evolved for mothers and daughters to stick together and to organize into snowflake-like clusters, which sink in liquid more rapidly than individual cells. This change in local population structure also reciprocally influenced the group's subsequent evolution: closely related groups of cells within one branch detached from the main cluster, resulting in a genetic bottleneck that greatly facilitated genetic segregation and subsequent evolution of larger clusters that settled faster (figure 5d).

The four examples above illustrate how even relatively simple selection scenarios can generate a feedback between spatial patterns and evolution in a diverse range of cellular systems: evolution tinkers with pattern formation which in turn can influence the driving forces of evolution.

5. Potential impact of higher-order biofilm patterns on the dynamics of evolution

Biofilms can exhibit a plethora of self-organized structural features, such as phenotypic stratification [8,70–72], channels [73] and wrinkles [8,74] (some of them shown in figure 1c). We believe that such higher-order structural features have greater influence on evolutionary dynamics than currently appreciated. Previous work has demonstrated the critical role of extracellular matrix (ECM) secretion for many of such structural features and has also explored the consequences of ECM production on the evolution of cooperation and competition [26,75]. For instance, ECM production can help producers to outcompete cheaters through spatial exclusion [76–78], or because the ECM can help retain public goods [79]. However, a systematic exploration of the link between the dynamic structure of biofilms and their evolution is still lacking. To illustrate potential avenues of further exploration we here outline two hypotheses for how the structural features of biofilms may influence their evolutionary dynamics and vice versa. Future experimental and theoretical studies will have to test these tentative suppositions.

Our first hypothesis is based on the observation that on solid agar substrates, ECM-producing biofilms have increased front undulations compared with non-producers (figure 6a) [43]. Based on previous work showing negative correlation of front roughness and establishment probability in colonies of E. coli and S. cerevisiae [34], we speculate that, via a similar mechanism, increased ECM production might hamper the success rate of de novo mutations in B. subtilis biofilms.

Figure 6.

Complex structural patterns in B. subtilis biofilms coupling to evolutionary dynamics. (a) Biofilms grown with inhibited production of extracellular matrix (ECM) have smoother fronts (i) than those secreting ECM (ii) with potential consequences for genetic drift [43]. (b) Side view of a surface wrinkle which is associated with movement out of the plane of population expansion [73]. (c) Wrinkles close to the front in B. subtilis biofilms are associated with front indentations (orange arrows). (d) Zoom of the indicated region in (c) showing that wrinkle initiation is preceded by a depression in the front. Images in (c) and (d) taken from still images of the supplementary video in [8]. (e) Hypothetical scenario of a slower-growing mutant (dark grey) causing a front depression (top left) giving rise to inward motion of surrounding wild-type cells (light grey) (top and bottom left). The ensuing mechanical frustration is resolved by buckling into the third dimension (bottom right) and wrinkle formation (shaded region). The associated additional retardation of the local front results in a positive feedback aggravating selective pressure on the mutant (top right).

As a second hypothesis we argue that there may be a positive feedback between surface wrinkle formation (figure 6b) and fitness heterogeneities within the population. The formation of wrinkles is closely related to ECM secretion [79,80] and is driven by cell-death-mediated out-of-plane buckling [74]. The associated redirection of growth-induced forces from in-plane to out-of-plane movement can have a detrimental effect on the local front propagation, as indicated by depressions in the front at the location of wrinkles (figure 6c,d) [8]. While such front undulations may occur spontaneously in some cases, they can also be facilitated by the presence of a slower-growing clone (figure 3b) [24,59]. Pending further investigation, we therefore propose the following hypothetical scenario, outlined in figure 6e: The front depression induced by a slower-growing clone causes an inward lateral motion of the adjacent sections of the biofilm. The resulting mechanical frustration triggers local buckling succeeded by wrinkle formation. The presence of a wrinkle, in turn, feeds back on subsequent growth, further retarding local front propagation. Such a mechanism could potentially lead to a more effective purging of slower-growing clones from biofilm populations. Similar to the mutation rate [81,82], purging efficiency may in principle be a trait that can be subject to selection, for instance, to decrease the genetic load of the population. However, it could also be an unavoidable consequence of the crowded nature of these biofilms.

6. Towards a systems biology of populations

We have argued that it would be worthwhile to study quantitatively how evolutionary forces emerge from the stochastic and deterministic behaviour of individual cells. The significance of this endeavour can be appreciated by comparing the current treatment of ‘noise’ in the fields of evolution and systems biology. Both fields accept the importance of noise: in systems biology, detailed measurements of how noise emerges within cells have enabled refined models of stochastic gene expression, which now form an essential pillar of systems biology [83]. Evolutionary biology also embraces noise in the form of random genetic drift, which controls the fate of newly arising mutations and hence the pace of evolution. However, because few quantitative measurements of genetic drift exist, models are usually based on generic well-mixed models. These models ignore spatio-temporal correlations, which are a hallmark of self-organizing systems. A truly predictive understanding of evolutionary dynamics requires, as in systems biology, empirically grounded models of the emergent phenomena, thereby accounting for spatio-temporal correlations and collective effects. An important step in this direction are individual-based computer simulations that begin to take into account the discreteness of individuals as well as the spatial degrees of freedom [34,57,84,85]. But a whole orchestra of cell–cell interactions is waiting to be explored, including steric interactions, the third dimension, friction, the viscoelasticity of the extracellular matrix and hydrodynamics. The difficulty of simulating every detail, and the challenge to interpret those complex simulations, will make it essential to compare computational approaches with predictive mathematical models that incorporate only few dynamical features.

Just how the paradigm shift of noise in cellular biology required input from other disciplines of science, this will also be required for higher levels of organization. In particular, the task of bridging the scales from small to large could truly benefit from statistical mechanics, which has long dealt with emergent phenomena in complex systems exhibiting many degrees of freedom. And, as with noise in cell biology, it may take several years to establish that pattern formation is not just a generic feature of population biology but key to the dynamics of evolution. Yet, we are confident that it is well worth the effort as it will greatly advance our understanding of the evolutionary principles governing natural populations.

Data accessibility

This article has no additional data.

Competing interests

We declare we have no competing interests.

Funding

This work was supported by a grant from the Simons Foundation (#327934, O.H.), by a National Science Foundation Career Award (#1555330, O.H.) and a NIH grant (R01GM115851). J.K. acknowledges a Research Scholarship (KA 4486/1-1) awarded by the German Research Foundation (DFG). Q.Y. was supported by the National Science Foundation Graduate Research Fellowship under grant no. (DGE 1106400) and a Chancellor's Fellowship awarded by the University of California, Berkeley.