Observation (1) asserts that the behavioral expression (phenotype) of an organism is influenced by exogenous inputs, in addition to genetic (i.e. hereditary) determinants. This flexible response to exogenous factors is referred to as phenotypic plasticity. The plasticity of interest to Baldwin, a psychologist, was cognitive flexibility based on the capacity to learn. This capacity has genetic underpinnings, but the content acquired through learning is generated through interaction with the environment and, although quite useful to the learner, is not passed on by genetic transmission. Observation (2) asserts that genetic mutations can occasionally give rise to phenotypic expressions equal or similar to those in (1) that require exogenous inputs. Advantageous characteristics that arise through plasticity need to be acquired afresh in each generation. This might incur costs, delays, and outcome uncertainty. If these same characteristics were to arise at the genetic level, they would be certain, possibly less costly, and directly selectable, which leads to observation (3) that such genetically encoded characteristics will sweep the population.
Overall, the process (1)-(3) makes it appear as if an advantageous but non-hereditary characteristic has become hereditary in some dangerously Lamarckian-looking fashion – an environmentally induced or learned behavior has become automatic. Baldwin, Morgan, and Osborn point out that the Lamarckian flavor is only an illusion; the crux of the matter is that plasticity enables a population of organisms to persist for prolonged times in a new environment. Continued residence in the new environment is not only advantageous, but exposes the population to precisely those selection pressures that reward the genetic “hardwiring” of a characteristic coincident with the inducible one. The argument is perhaps best clarified by means of a concrete example drawing on a much simpler form of plasticity with a mechanistic basis.
An animal finds better food at high altitude. There is less oxygen at high altitude, but the animal adapts plastically to lower oxygen levels. In response to lower oxygen, the animal generates 2,3-diphosphoglycerate, DPG, a molecule that lowers the oxygen affinity of hemoglobin, causing it to unload oxygen more thoroughly. The new behavior of hemoglobin induced by an environmental condition permits the population to linger in a hostile environment that provides a benefit (better food). This, in turn, gives selection pressures a chance to reward mutations of hemoglobin that do on their own what current hemoglobin can only do in the presence of DPG. Hence, in a first phase of the process, animals that adapt plastically to higher altitudes (by invoking suitable chemistry) win over those that can’t. But doing so requires a costly molecular apparatus with many “moving parts” each of which can cause problems. A form of hemoglobin with “hardwired” high-altitude characteristics – independent of supporting chemistry – would be a really good idea. Thus, in a second phase, those individuals that acquire a mutation producing such a version of hemoglobin from the get go will have a selective advantage over those who need to ramp up supporting chemistry. The net effect is that a flexible characteristic facilitated adaptive evolution in the direction provided by that flexibility. This is the Baldwin effect. The idea can be iterated: Should flexibility itself be the desired characteristic, the same process can be invoked for flexibility to acquire a genetic (heritable) basis.
It is important to understand that the Baldwin effect does not assert that a favorable mutation is more likely to arise; it only asserts a process that facilitates evolution by permitting specific selection pressures to become effective. It gets more intriguing.
The future is now
For a brief moment, Simpson entertains the idea of a causal connection between (1) and (2), but he reaches the conclusion that any nexus between the plasticity of a biological system and the likelihood of a favorable mutation must be Lamarckian: The adaptive response of a biological system to an environmental situation somehow causes an appropriate mutation within the same individual during its lifetime. Naturally, Simpson dismisses this possibility.
Conrad H. Waddington replies that Simpson has overlooked something: genotypes with “the ability to produce an adaptive phenotype would […] encourage the appearance of genetically controlled variants [i.e. mutations] mimicking the adaptive type”. This needs some work; back to the hemoglobin story.
A protein, such as hemoglobin, is a sequence of chemical beads folding up into a native three-dimensional structure that conveys chemical and biological “function”. More precisely, a string of chemical beads can fold up into many forms, but (for the sake of argument) only the most stable one is adopted spontaneously; call it the “typical” form. Other forms can be adopted too, but they require some prodding from other molecules, such as DPG. These alternative forms reveal themselves only in response to such exogenous triggers. Imagine next that the sequence-to-shape relationship of proteins has the following generic properties (*):
The significance of properties (1) and (2) for evolutionary change can be readily appreciated. If selection could evaluate a present individual on the basis of the characteristics of its future offspring, evolution might speed up considerably. But selection has no foresight. However, if the present were an indication of the future, selection would act as if it had foresight. This is precisely the case when the possible alternative structures of a protein sequence are those that can be easily stabilized into native structures upon mutation. Properties (1) and (2) posit a kind of congruence: a correspondence between plasticity – here defined as the capacity to attain non-native behaviors through sustained interaction with an environment – and variability – here defined as the capacity to access new native behaviors (**). The net effect of such congruence is that selection, in evaluating a single point X in genotype space, automatically evaluates a genetic neighborhood of X, because its native behaviors are already manifest as inducible behaviors of X. Provided the environment samples those behaviors, this can lead to a dramatic speed-up in the rate of evolution. Properties (1) and (2) are an idealization. An advantageous alternative form does not have to become native in one single mutational step; rather, it may gradually increase in dominance within the plastic range, becoming native only gradually after several mutations – a process Waddington called “genetic assimilation”. In 2000, Lauren Ancel and I established computationally the existence of a statistical congruence of this kind for RNA sequences and their structures.
The correspondence between behaviors that are endogenous and those that require exogenous assistance adds something fundamental to Baldwin: The particular mechanics of the mapping from protein or RNA sequences to structures makes it vastly more likely that a mutation tosses up a sequence with a favorable native structure once evolution has hit upon a sequence with a favorable inducible structure.
For Waddington, mechanisms that promote plasticity automatically possess an exploitable degree of congruence. I wonder whether his suggestion corresponds to a general insight from the theory of dynamical systems: Those directions in which an attractor (a stable dynamical behavior) changes the most when parameters are being varied (think mutation) correlate with the directions that are slowest in restoring the unperturbed state in response to perturbations of the dynamical variables (think plasticity).
Facilitated variation
For Waddington, a genotype was a range of phenotypic possibility in a particular environment. That range becomes manifest through plasticity, i.e., the capacity to adapt in a non-hereditary fashion through learning or physical interaction with the environment. A genotype is selected on the basis of that range, and the effect of a mutation consists in modifying that range. While a mutation may be completely random as to when and where it occurs, its consequences on the phenotype (the plastic range of possibility) are far from random. A phenotype is the outcome of a process that relies on some mechanism, and a mechanism necessarily biases in specific ways the consequences of its modification. Two distinct mechanisms that generate the same behavioral repertoire need not (and typically are not) equivalent with regard to their potential for change.Waddington refers to this point as “unduly neglected by neo-Darwinism”.
Waddington’s line of argument was focused on how mechanisms that enable a range of phenotypic possibility (plasticity) may automatically provide an opportunity for small modifications to “canalize” a feature within that range. He appears less concerned with whether (or how) plasticity promotes novelty.
Marc Kirschner and John Gerhart pushed Waddington’s train of thought into a new direction informed by modern molecular developmental biology. They refocus the issue on mechanisms that enhance the effectiveness of evolution in generating novelty. First, Kirschner and Gerhart generalize Waddington’s scenario to multiple nested levels of biological organization within an organism. For example, a cell does not only experience an environment external to the organism, but also an environment generated within the organism by other cells. Second, they suggest specific molecular and architectural design principles that provide a mechanistic foundation for the facilitation of evolutionary innovation. Briefly:
To appreciate the new angle with respect to Waddington, it might help to think of a property as being implemented by a mechanism f(e,i) that depends on exogenous and endogenous parameters e and i, respectively. In the hemoglobin example, i is the amino acid sequence and e is the DPG molecule. (The system of interest was hemoglobin, not the organism as a whole. Thus DPG is exogenous with respect to the hemoglobin sequence.) Waddington’s scenario leads to a transfer of causality, and thus control, from e to i. Congruence is the phenomenon that facilitates this transfer. In the case of RNA or protein molecules, congruence is not particularly mysterious, since both exogenous and endogenous parameters influence the same free energy landscape of folding. Any chemical interaction will do, regardless of whether it occurs entirely within the sequence (genetically encoded) or between the sequence and another molecule (inducible): Such interactions are fungible.
In this picture, Kirschner and Gerhart shift the emphasis towards a variation in f (the heritable mechanism) itself. A modification of f, too, could remove the dependency of an advantageous property on an exogenous e. This is hard to imagine for hemoglobin, where f is “the physics of folding”, which is not encoded by genetic factors. Variation of f is easier to envision when it is genetically encoded, as is the case for a network of molecular signaling components. The question then becomes whether an architecture that enables f(e) to be plastic (i.e. responsive to exogenous signals) automatically permits f to be easily modified while remaining functional. Recall that Waddington’s scenario posits an alignment between the facultative phenotypes of a parent genotype and the obligatory phenotypes of its mutant offspring, not between the facultative behaviors of parent and offspring. In Kirschner ‘s and Gerhart’s scenario, novelty arises because of the appearance of new facultative (inducible) behaviors as genotypes (encoding f) are mutated.
The distinction between varying e and varying f may not always be easy to maintain. A similar situation arises for the concepts of “program” (analogous to f) and “data” (analogous to e): A universal Turing machine may produce any behavior, because the processing of data can amount to executing a program encoded in the data. However, the weak linkage principle suggests that a distinction between e and f can be maintained, because signals that trigger complex responses are typically of low complexity. Then, the architecture of f, acquired while evolving to flexibly respond to a dynamic environment e, is automatically such that its modifications are more likely to successfully exploit environmental situations outside the “evolutionary training set” of f.
I'm looking for postdocs or risk-taking graduate students to work on these problems, utilizing the theoretical/computational framework we developed to study molecular signaling.wf
(*) Generic here means that these properties hold statistically for “most” sequences.
(**) In pseudo-formal speak, plasticity might be defined as the change of phenotype P as a function of the environment E at constant genotype G (the derivative of P with respect to E at constant G); whereas variability might be defined as the change of P as a function of G at constant E (the derivative of P with respect to G at constant E).
