6
The prior chapter summarized the forces that change fitness value. Fitness value considers traits abstractly. This chapter reviews how traits develop, what causes traits to vary, and where new traits come from.
The first section contrasts abstract and mechanistic aspects of traits. Abstractly, changed genetic mixing predicts a change in growth rate. That abstract prediction provides broad insight. But it also ignores the mechanistic basis of growth.
Mechanistically, growth depends on the underlying biochemistry and biophysics. The fundamental forces of value shape design through mechanistic change. Mechanistic insight improves the accuracy of comparative predictions and broadens the understanding of design.
The second section discusses the modification of traits. In some cases, small changes to existing traits may be sufficient. Attack less. Disperse more. Heritable quantitative variation often exists, providing the basis to adjust traits.
Big environmental shifts may require large changes in traits. Large variants may not exist. To meet that challenge, the processes that generate variation may evolve. Increased mutation, genomic rearrangement, and genetic mixing generate greater variation. Generative processes modify the evolutionary rate of traits.
The third section considers the origin of traits. How do cells acquire resistance to a novel toxin? How can a cell switch to a novel food source?
In the first step, a novel genotype may arise. However, complex traits often require the simultaneous evolution of several components. A single genetic novelty by itself may be of little value.
Alternatively, the path to a novel trait may begin with a phenotypic variant of a common genotype. The initial phenotypic variant may not produce the favored trait. But it can bring a genotype closer to the favored form.
With a partial solution from an initial phenotypic variant, subsequent genetic variants can more easily transit to a novel character. This phenotypes-first sequence greatly accelerates evolutionary discovery.
Organisms often plastically adjust phenotypes to match the environment. Because plasticity typically covaries several mechanistic components, plasticity may generate variety in the right direction with regard to a novel challenge. Genetic variation and selection can then modulate the initial variety, steadily moving toward the favored form.
6.1 Nature of Traits
More genetic mixing reduces similarity between neighbors, which enhances growth rate to outcompete genetically distinct neighbors. In particular,
genetic mixing ⊣ similarity ⊣ growth rate,
which expresses an interesting and testable comparative prediction. This prediction considers growth rate abstractly, ignoring the mechanisms that determine the trait.
What determines growth rate? Genes do not encode growth rate. Instead, genes influence the expression of molecules, which alter the uptake of substrates and the sensing of food concentrations. Nucleotide sequences affect the binding kinetics of transcription factors, which trigger switching between metabolic pathways.
Do these mechanisms matter when trying to understand the forces that shape growth rate? At one level, they do not. The comparative prediction for genetic mixing typically holds for different mechanistic assumptions. Not always, but likely often enough that one expects the predicted direction of change in growth to happen more often than not.
At another level, the mechanistic basis of traits provides deep insight into the forces that shape design. Consider two contrasting mechanisms that influence growth rate.
First, a gene duplication may increase the expression of a cell surface transporter that pulls sugar into the cell. Greater uptake rate for sugar may enhance growth rate. Making more transporters requires additional resources, reducing the efficiency yield at which a unit of sugar is transformed into a unit of reproductive biomass. This mechanism creates a tradeoff between growth rate and reproductive yield.
Second, a nucleotide substitution in an enhancer of gene expression may trigger a faster switch of metabolism between alternative sugars. That faster switch reduces the variance in growth rate by speeding metabolic transitions when conditions change.
Mechanism provides new predictions. Suppose, for example, that increased resources favor high growth rate at the expense of reduced yield. If transporter duplications alter substrate uptake, then enhanced growth may be mediated by the gain of duplicated transporter genes.
Alternatively, suppose that fluctuating conditions favor mechanisms to reduce the variance in growth at the expense of lower average growth. Modified enhancers may reduce grow rate variance by speeding the switch between alternative food sources. The costs for mechanisms of fast switching may reduce overall average growth rate.
In this book, I present many abstract, mechanism-free predictions. Those abstract predictions are simple, general, and broadly applicable.
I also develop many predictions that depend on mechanism. Those mechanism-based predictions provide essential insight into the design of traits. We need both abstract and mechanistic perspectives to enhance our understanding of design.
6.2 Modification of Traits
Comparative predictions forecast the direction of change in traits. Often, we focus on quantitative changes. Grow faster. Secrete less.
When the change is small, heritable variation typically exists or arises de novo. Natural selection can often make small quantitative adjustments in traits.
Large environmental shifts create strong forces, which may favor significantly changed traits. Big changes in traits may depend on enhanced generative processes to provide new sources of variation,
large environmental shift → strong forces → generative processes.
This prediction considers generative processes, including genetic mutation and genomic rearrangement, as traits shaped by the forces of design. A generative process functions by modifying the evolutionary rate of other traits.308,311
6.3 Origin of Traits
Upon exposure to a novel toxin, resistance may require a novel mechanism. By what evolutionary sequence does a new resistance trait arise? In general, how do new traits evolve?
GENES FIRST
Perhaps novel genotypes arise by chance. A new genotype may create a new trait or qualitatively alter an existing trait. For example, a new genotype may produce a novel antitoxin or significantly alter an existing antitoxin. First, the genotype arises by chance. Then selection of the genetic variant follows.
The range of traits produced by genetic variants depends on the physical basis by which phenotypic variants arise. For example, if a novel antitoxin requires only a change in the external binding site that attaches to the toxin, then such novelty may arise relatively easily.
By contrast, if existing antitoxins lack the required mechanisms to neutralize a newly encountered toxin, then simply modifying the binding properties of existing antitoxins is not sufficient. Both novel binding and neutralization aspects may be required. Such novelty may rarely arise by just a few simple genetic changes.
The genes-first pathway to novelty has been widely discussed in evolutionary theory.305 The remainder of this section focuses on an alternative pathway to novelty that has received less attention.
PHENOTYPES FIRST
Perhaps a novel phenotype first appears by variant trait expression among individuals with a shared genetic basis for the trait. Eventually, new genetic variants may heritably stabilize the favored phenotype.23,132,142,183,268,270,358,431,432,443,451
For example, cells may use generic pumps to excrete toxins from the cell. Toxin pumping may vary stochastically between cells because pumps depend on a small number of intracellular molecules. Upon initial challenge by a toxin, the survivors may be those phenotypic variants that, by chance, highly express toxin pumps.
Among the survivors, subsequently arising genetic variants may upregulate toxin pump expression, modifying the original trait. Increase of those new genetic variants permanently raises trait expression, stabilizing the favored change.
Cell division rate provides an alternative mechanistic pathway to increased resistance. Suppose the toxin works only against actively dividing cells. Cells vary stochastically in the time between cell division. Quiescent cells resist attack.
Among quiescent cells that survive, a descendant lineage may eventually gain a mutation for a novel resistance trait, such as a modified antitoxin or a variant cell-surface receptor. Increase of the new genetic variant stabilizes the favored change.
In general, a phenotype-first process to generate variability can greatly increase the rate at which traits evolve in response to strong environmental challenge. The evolutionary response may modify an existing trait or create a novel trait.
STOCHASTICITY SMOOTHS THE FITNESS LANDSCAPE
Phenotype-first variation accelerates evolutionary discovery by smoothing the fitness landscape.132 Modification of an existing trait illustrates the theory. The same principles apply to the origin of new traits.
Suppose that each genotype produces an average trait value, μ. The value of μ varies between genotypes. We can write the probability distribution of phenotypic expression for a given genotype as p(x|μ), the probability of observing a phenotypic value of x for a genotype with mean value μ. In the following examples, I assume a normal distribution with variance γ2 for all genotypes.
The top row of Fig. 6.1 shows the distribution of phenotypes for a genotype with mean value μ. The solid curve traces a distribution with a relatively small variance. The dashed curve follows a distribution with a relatively large variance.
The second row in that figure shows the fitness, f(x), associated with each phenotype, x. On the left, fitness is high only when the phenotype is very close to the optimum. Other phenotypic values have zero fitness.
The third row shows the average fitness, F(μ), of a genotype with mean phenotype, μ. The average fitness weights each fitness value, f(x), by the probability, p(x|μ), of expressing the phenotypic value, x, as
This transformation for fitness begins with the initial fitness landscape that associates a phenotype, x, with a fitness value, f(x). The distribution of phenotypes, p(x|μ), expressed by each genotypic value, μ, acts as a smoothing filter to produce the final fitness landscape, F(μ).
The transformed fitness landscape in the lower left of Fig. 6.1 illustrates the smoothing process. The original fitness landscape, f(x), in the panel above is a narrow peaked function. To survive and obtain nonzero fitness, a phenotype must almost exactly match a specific expression.
Figure 6.1 Variable phenotypes and fitness. Each column shows how the distribution of phenotypes expressed by a given genotype, p(x|μ), smooths the fitness function, f(x), to give the expected fitness, F(μ), for a genotype with average phenotype μ. The smoothing follows eqn 6.1. These examples use normal distributions, ?(μ, δ2). The distribution p has variance δ2 = γ2, the shape of f follows a curve with variance σ2, and F follows a curve with variance γ2 + σ2 (see Frank132). (a) The solid and dashed curves show the phenotypic expression, p(x|μ),which follows ?(μ, 1/2) and ?(μ, 5), respectively. Fitness, f(x), has the shape of a normal distribution with vanishingly small variance, ?(0, σ2→0). Thus, expected fitness, F(μ), is the same as the phenotypic expression, p. (b) The same structure as in (a), except that f(x) is much wider, following ?(0, 7). Thus, F(μ) now follows ?(0, 7.5) and ?(0, 12) for solid and dashed curves, respectively. In each plot, the baseline is set to 4.3% of the peak in that plot. The baseline truncates phenotypes with low vigor, setting their fitnesses to zero. From Figure 2 of Frank.132
If each genotype produces a particular phenotype without any variation, then, using Maynard Smith’s268 language, matching a genotype to the favored form would be like searching for a needle in a haystack. By contrast, if each genotype produces a distribution of phenotypes, then, as in the bottom row, matching a genotype to the favored form would be like searching for a needle in a haystack when someone tells you when you are getting close.
Figure 6.2 A broad expression of phenotypes smooths a multipeak fitness landscape. (a) The dashed curve shows broader phenotypic expression, p(x|μ). (b) The fitness landscape, f(x), has multiple peaks. (c) Broad phenotypic expression (dashed curve) smooths the realized fitness landscape, F(μ), to a single peak. In this example, the narrow and broad phenotypic expression patterns follow ?(0, γ2) with variances of 0.04 and 0.16, respectively. Fitness is given by , with σ2 = 0.0225. The value of F(μ) is calculated from eqn 6.1, yielding the expression for f(x) in the prior sentence with the variance replaced by σ2 + γ2. The baseline truncates small values. From Figure 4 of Frank.132
Put another way, phenotypic smoothing transforms the problem of exactly matching a target phenotype, without any clue about how close the current genotype is, into the problem of climbing a smooth fitness gradient to a local peak. Natural selection is very bad at finding special traits for which nearby phenotypes have low fitness. Natural selection is very good at improving traits by climbing a fitness gradient toward a local peak.
PEAK SHIFTS ON SMOOTHED LANDSCAPES
Multipeak fitness landscapes provide the classic model for the origin of new traits.62,455,456 Figure 6.2b illustrates a multipeak landscape. Suppose the phenotypes of a population cluster near the middle peak. That population cannot evolve toward the higher peak by climbing a fitness gradient.
Put another way, small quantitative modulations of the current trait cannot improve performance. Instead, a qualitatively distinctive shift in the trait may be required to achieve the higher peak.
How can a population discover the improved trait when small modifications of the existing trait reduce fitness? Stochasticity of phenotypic expression provides one solution. Stochasticity can smooth the fitness landscape, transforming the difficult peak-shift problem into the simple problem of climbing a smooth fitness gradient.132
Suppose the phenotypic expression for a given genotype varies. Figure 6.2a shows two examples for stochastically variable expression, p(x|μ), which is the distribution of phenotypes, x, for a genotype with mean phenotype, μ. Once again, we can apply eqn 6.1 to obtain the fitness landscape that matches each genotype’s mean phenotype, μ, to its fitness, F(μ).
Small amounts of phenotypic stochasticity, shown by the solid curves, partially smooth the landscape in Figure 6.2c. However, distinctive peaks remain. The population cannot climb a smooth gradient from the middle peak to the higher peak.
Greater phenotypic stochasticity (dashed curve) smooths the landscape into a continuously rising gradient toward a single peak. Natural selection can push the population up the smooth gradient to the peak.
Increased stochasticity transforms the difficult problem of shifting between peaks into the easy problem of climbing a hill. Phenotypic stochasticity accelerates the discovery of novel trait expression.
PHENOTYPIC PLASTICITY AND NEW TRAITS
Stochasticity accelerates evolutionary rate. However, stochasticity typically confines phenotypes within an existing set of trait values.
How can evolution discover qualitatively distinctive traits in response to environmental challenge? New genetic variants may produce novel traits. However, creating complex novel traits in one step may be difficult.
Instead, novel phenotypes may first arise when individuals respond to new environments by adjusting their development, physiology, or behavior. This phenotypes-first pathway for generating novelty may play an important role in the evolution of new traits.443
If novel traits created by phenotypic plasticity partially meet the new environmental challenge, then natural selection of genetic variants can favor quantitative improvements of those novel traits, climbing a local fitness gradient. This process transforms the difficult problem by which evolutionary process discovers novelty into the simple process by which natural selection climbs a hill to improve an existing trait.
Comparatively, greater phenotypic plasticity increases the rate at which novel traits evolve in response to environmental challenge. For example, the rate at which a microbe acquires resistance to a novel toxin or drug may increase with the microbe’s plastic response to environmental stress. Consider a speculative scenario to illustrate the process.
Suppose a microbe comes under attack by a novel toxin. The toxin enters the cell by binding to a cell-surface receptor. If the cell can survive for a period without the receptor, then shutting down receptor expression in response to the toxin may allow the cell to resist attack.
To shut down receptor expression, the cell must evolve a sensor for the toxin and the regulatory pathways that link the sensor to the expression of the receptor. Simultaneous evolution of coordinated components is difficult. The sensor alone provides no advantage. A regulatory switch provides no advantage without appropriate sensors.
Genetic mutations rarely provide such simultaneous jumps in multiple traits. A genes-first pathway for the origin of complex traits seems difficult. Put another way, no simple gradient of increasing fitness exists for natural selection to climb.
However, cells often have pre-existing regulatory pathways for responding to environmental cues. Such phenotypic plasticity greatly enhances the opportunity to evolve new traits. For example, a generic cellular response to attack may shut down the expression of several receptors, including partial reduction of the target receptor for the toxin.
The existing plasticity of the cell already creates a phenotype that is close to the required trait. In the next evolutionary step, a novel mutation may link an existing sensor, partially stimulated by the toxin, to the general defense response. That step would climb up the fitness gradient, in which a direct change in one character raises fitness.
Once a specific receptor is linked to the regulatory control that switches expression, further genetic variants that modulate the response may be favored by climbing the hill of increasing fitness.
In this scenario, the evolutionary path started with a phenotypic variant induced by the cell’s intrinsic phenotypic plasticity. Then genetic variants in sensors and regulatory wiring improve the initial phenotype. The difficult problem of novel trait discovery transforms into the simple problem of continual improvement by small changes.
PHENOTYPES-FIRST PATHWAY ACCELERATES EVOLUTION
A genotypes-first pathway for new traits is easy to understand. Mutations arise randomly. Large populations contain many mutants. A new environmental challenge favors one of the pre-existing mutant genotypes.
The genotypes-first pathway suffers from a major difficulty. Selection can favor only those traits that originate by genetic changes, which we may broadly call mutations. Discovering novel traits by mutation may not happen easily. For example, how do two mutually beneficial traits arise when neither one alone provides value?
A phenotypes-first pathway can accelerate trait discovery. Each existing genotype produces a range of phenotypes. The first step requires only that a genotype express a trait that gains a little bit with respect to a novel challenge. That first favored form arises by nonheritable phenotypic variability.
Subsequently, selection favors those genotypes that can produce phenotypes more closely matched to the target. The difficult problem of discovery transforms into the relatively simple problem of hill climbing.
Phenotypes first does not automatically discover two synergistic traits. But it makes discovery easier because plasticity often responds to environmental challenge by modulating suites of interacting mechanisms.
By covarying mechanistic components, plasticity generates phenotypic variety that may be in the right direction with respect to a novel challenge.443 That phenotypes-first step toward discovery smooths the fitness landscape. A smoothed landscape provides a more direct path of genotype change, accelerating the process of novel trait evolution.
Despite the long history and extensive literature on phenotypes-first trait discovery, this aspect of adaptation remains an underappreciated evolutionary force. Comparatively, greater stochasticity of trait expression or greater phenotypic plasticity increases the rate of novel trait discovery, mediated by the smoothing of the fitness landscape.