A principle of organization which facilitates broad Lamarckian-like adaptations by improvisation
During the lifetime of an organism, every individual encounters many combinations of diverse changes in the somatic genome, epigenome and microbiome. This gives rise to many novel combinations of internal failures which are unique to each individual. How any individual can tolerate this high load of new, individual-specific scenarios of failure is not clear. While stress-induced plasticity and hidden variation have been proposed as potential mechanisms of tolerance, the main conceptual problem remains unaddressed, namely: how largely non-beneficial random variation can be rapidly and safely organized into net benefits to every individual.
We propose an organizational principle which explains how every individual can alleviate a high load of novel stressful scenarios using many random variations in flexible and inherently less harmful traits. Random changes which happen to reduce stress, benefit the organism and decrease the drive for additional changes. This adaptation (termed ‘Adaptive Improvisation’) can be further enhanced, propagated, stabilized and memorized when beneficial changes reinforce themselves by auto-regulatory mechanisms. This principle implicates stress not only in driving diverse variations in cells tissues and organs, but also in organizing these variations into adaptive outcomes. Specific (but not exclusive) examples include stress reduction by rapid exchange of mobile genetic elements (or exosomes) in unicellular, and rapid changes in the symbiotic microorganisms of animals. In all cases, adaptive changes can be transmitted across generations, allowing rapid improvement and assimilation in a few generations.
The hypothesis raises a critical, but thus far overlooked adaptation problem and explains how random variation can self-organize to confer a wide range of individual-specific adaptations beyond the existing outcomes of natural selection. It portrays gene regulation as an inseparable synergy between natural selection and adaptation by improvisation. The latter provides a basis for Lamarckian adaptation that is not limited to a specific mechanism and readily accounts for the remarkable resistance of tumors to treatment.
It is commonly accepted that stable adaption to new environments is mediated mainly by natural selection of individuals carrying advantageous genetic changes. This view distinguishes the long term evolutionary process of genetic adaptation from physiological adaptation that may take place within generation. The latter is often viewed as ‘execution’ of previously evolved programs. This strategy of physiological adaptation is mostly effective for coping with conditions that have been encountered repeatedly during the evolutionary history of the organism (e.g. exposures to heat, starvation, drought, predation, etc.). However, in the case of rare or completely novel scenarios of stress, the organism is unlikely to have had sufficient prior opportunities to evolve specific programs of adaptation. Whether and how organisms can still mount adaptive responses in these cases is largely unknown.
The requirement for an ability to cope with novel scenarios of stress is not limited to cases of evolutionarily novel environments (which may indeed be rare). Even in highly regular environments, every individual is expected to encounter many individual-specific combinations of internal changes which can compromise the otherwise beneficial outcomes of pre-evolved mechanisms. These include, for example, somatic genetic aberrations in every cell, epigenetic changes in cells, microbial changes within the organism, etc. Additionally, changes within a cell of a particular tissue may impact other cells within and outside this tissue, thus leading to co-existing aberrations at multiple scales (from intracellular process to the whole organism and its environment). Since the combinatorial space of these changes is immense, an overwhelming fraction of the specific combinations of changes could not have been sampled during evolution. Of course, many of the non-sampled (i.e. novel) changes can be dealt with by highly efficient, pre-evolved generic mechanisms. For example, genetic adducts are efficiently removed by DNA repair mechanisms [1], aberrant cells are eliminated by apoptosis [2, 3], immune surveillance [4–8] and/or additional mechanisms [9], and pathogenic micro-organisms are contained or eliminated by the immune system [10–13]. However, these generic mechanisms cannot provide complete protection against all scenarios of disrupted homeostasis [14–26]. For example, the immune system successfully eliminates many viruses, but some of them manage to evade this system using a variety of approaches [27–31]. Such viruses have been demonstrated to substantially alter transcription in cells of the infected host [32, 33] and can generate novel functional transcripts [34–37].
Since almost all living organisms contain a multitude of viruses which modify the host and evolve at a greater pace than the host genome [38–41], there can be no genetic program for handling all the possible scenarios of maladaptive conditions. A similar limitation applies to coping with individual-specific epigenetic drifts associated with aging [42–44] and recovering from other types of individual-specific lesions (e.g. microinfarcts in various tissues [45, 46]). This shortcoming is reflected in many cases of dysfunction that have individual-specific components, such as chronic illnesses, auto-immunity and tumorigenic transformations.
It is generally accepted that coping with novel scenarios of severe stress requires phenotypic plasticity [47–56] and may involve exploratory processes [56, 57]. However, without mechanisms that have been previously selected based on their ability to mount beneficial responses to individual-specific conditions, it is not clear how plastic changes and exploratory processes can be adaptive. In fact, in the absence of efficient ways to bias plastic changes towards beneficial outcomes, plasticity is expected to do more harm than good (much like random mutations). This problem cannot be addressed by natural selection, which operates at the (large) population level, but is ineffective when each individual encounters a large number of individual-specific perturbations ('failures') due to internal changes. Put differently, mutations which could protect some individuals under rare external stress do not assist in coping with a multitude of individual-specific combinations of internal failures. Phenotypic robustness is also not effective enough because robust phenotypes may become maladaptive in a non-negligible fraction of the internal failures. A qualitatively different strategy of (‘quasi-Lamarckian’) adaptation to novel environment [58, 59] has recently been inspired by experimental evidence of stress-induced mutagenesis [60–68]. It proposed that environmental conditions trigger non-specific mutations which confer adaptation to the stress-inducing factors [59]. The deleterious potential of such random mutagenesis was thought to be addressed (in prokaryotes) by regulation which involves DNA repair and particular structure of genome architecture. This regulation is presumed to bias mutations in a way that allows coordinated evolvability of functionally linked genes in rare cells where beneficial mutations emerge [59].
While this could indeed benefit rare cells in novel environments, it does not address the conceptual problem that is mentioned above, namely: how any regulation could use mostly non-beneficial variation to provide every individual with some capacity to make beneficial use of this variation in arbitrary types of novel conditions. Addressing this problem becomes far more challenging when each multicellular individual has to cope with many (individual-specific) combinations of internal perturbations in different cells, tissues and organs. Notably, the main question behind this adaptation problem is not how stress induces variation (or plasticity), nor how this variation can be transmitted across generations and whether it can be beneficial to some individuals under rare novel environments. The question is what kind of physiological regulation could facilitate de novo emergence of many multi-scale adaptations using random processes that are largely non-beneficial.
The importance of this question cannot be over-estimated. If every individual encounters many new scenarios of internal failures due to combinations of genetic and non-genetic changes, it is plausible that no individual can exist without some capacity to organize random variation in a beneficial manner. Lack of it may therefore prevent a population from even forming, thus invalidating natural selection altogether. This capacity, however, does not (and cannot) replace mechanisms that were established by natural selection, but rather complement these mechanisms.
Any conceptual mechanism which makes beneficial use of random changes must address the following critical questions:
How the induction of random changes could avoid disrupting essential processes which took a very long time to evolve?
How beneficial random changes could be identified within the immensely large space of non-beneficial changes? Here, we should bear in mind the relatively short time available to find viable solutions (less than one generation time) and the probable necessity of testing many putative solutions.
How random changes in different cells, tissues and organs can be coordinated into beneficial outcomes for the whole organism? For example, in the case of animals (as opposed to single cell organisms), it is not enough that every stressed cell will ‘find’ its own solution, because this may lead to non-coordinated (tumor-like) changes that severely compromise the organism as a whole.
In addition to these critical questions, it would be desirable (albeit not necessary) to propose a conceptual mechanism which is compatible with transgenerational inheritance of some of the newly-formed beneficial changes. Such inheritance would enable progressive improvement of the adaptation over generations. This should not be confused with examples of Lamarckian adaptation by pre-evolved mechanisms which address a specific type of novelty. Hallmark examples of these Lamarckian mechanisms are the CRISPR system in prokaryotes [69–71] and the small RNA-based mechanism of viral immunity in C. elegans [72]. Both mechanisms enable rapid acquisition of heritable resistance to new viruses, but the tolerance is always acquired using the same mechanistic pathway. While the preexistence of such pathway is very instrumental for coping with new viruses, it does not provide solutions for all other types of novel conditions (including pathogenic outcomes of new viruses that have managed to evade rapid elimination).
Potential signatures of a capacity to use random variation for generating individual-specific adaptations may be recognized in diverse contexts: Adaptive immunity in human incorporates ongoing generation of genetic changes in ways which permit adjustments of responses to co-evolving pathogens during the lifetime of a single person [73]. Similarly, the primate brain exploits neural learning for coping with new pathophysiological and intellectual problems. A striking example for this is given by de novo reorganizations of cortical motor neuron activities in ways which enable acquisition of control over a prosthetic arm [74]. Electrical stimulation of neural networks in a dish reveals analogous capacity for de novo learning in arbitrary configurations of stimulated neurons [75, 76]. A classic example in a developmental context was provided by a two-legged goat, born with a congenital paralysis preventing the use of the front legs [77]. This condition led to re-organization of anatomical features enabling hopping on the hind limbs [47, 53]. Functional reorganization of developmental processes is also apparent in various cases of mating between different species or breeds, in which distinct ‘programs’ are merged into functional outcome without a specific 'program of merging'. These cases include heterosis [78], plant grafting [79], mating of pure-bred dogs which differ substantially in their skull size and shape ([80], page 556) and even cross-genus cloning of one species into another [81]. The capacity to form new adaptations within one or few generations is not at all limited to multicellular organisms. It was clearly demonstrated by synthetic gene recruitment in yeast which de-coupled an essential gene (His3) from its endogenous regulation and placed it under a non-related promoter (GAL4).
Repression of the GAL4 promoter by switching to glucose-based medium drove rapid adaptation [82] which varied substantially between replicated experiments [83, 84] and did not necessarily involve genetic changes [85]. An additional mass of supportive (though less explicit) evidence in non-engineered settings of micro-organisms is provided by rapid acquisition of non-coding DNA and mobile genetic elements in bacteria [86–89], which may account for their surprisingly high rate of acquisition of anti-viral and antibiotic resistance [59, 87–89].
Coping with novel scenarios of stress is enabled by a large number of readily changing (i.e. ‘flexible’) traits or processes (collectively referred to as traits or features), combined with a sufficiently strong inverse-correlation between the flexibility of traits and their deleterious potential. Adaptation is then achieved by occasional stress reduction due to random changes primarily in those flexible traits and processes (‘Adaptive Improvisation’) and is further enhanced and stabilized by auto-regulation. Herein, we operationally define adaptation (or beneficial variation) as any change in the organism which reduces stress. Without a sufficiently strong inverse-correlation between flexibility and deleterious potential, the probability of deleterious outcomes would have been too high because most of the random variation is not beneficial.
Wide distribution of flexibilities: The traits and processes of each individual cell and organism have different capacities to vary in response to external or internal perturbations. Some traits are tightly regulated and are kept within a small margin of change (stable or ‘strongly constrained’ traits). Others are much more ‘flexible’ (weakly constrained) and can vary significantly and rapidly during the lifetime of the individual. These traits can include: a) expression, activity, location and affinity of non-essential genes or regulatory RNA, b) conformations of intrinsically unstructured proteins, c) abundance and exchange of non-essential mobile elements, and d) any other type of change that is not constrained by tight regulation. We assume that the distribution of flexibilities is very wide, so that every individual has a very large space of flexible features. A qualitative example of a putative distribution is illustrated in Fig. 1.
Inverse connection between flexibility and deleterious potential: A change in a flexible trait tends to be less detrimental compared to similar magnitude of change in a highly constrained trait (Fig. 1). This assumption is a direct outcome of natural selection, which filters out those individuals in which deleterious changes are not prevented by stringent regulation.
Reciprocal feedback between stress and random changes: Novel conditions induce stress which drives modifications in various traits. The probability of inducing random variation in a particular trait increases with the degree of flexibility of this trait and the strength of the stress. Consequently, the stress is inducing changes primarily in flexible traits [90]. Combined with assumption 2, these changes cause less harm than a similar magnitude of change in more constrained feature.
Based on the above assumptions, we propose a general process of emergent adaptations by improvisation. This process can involve any type of mechanism, thus providing broad adaptive capacity which is not exclusively dependent on a specific mechanism. The core component of this adaptation is the alleviation of stress by drive reduction. While this might be sufficient for adaptation by improvisation, the outcome of it can be enhanced and stabilized by auto-regulation.
Any stress which is not sufficiently alleviated by existing (pre-evolved) stress responses drives rapid (not necessarily genetic) changes mostly in flexible and relatively non-harmful features. These changes can potentially alleviate the stress, but can also aggravate it. However, changes in directions which alleviate the stress (operationally defined as adaptive/beneficial), also reduce the drive for additional changes, thus promoting higher persistence of states (and processes) associated with lower stress (Fig. 2a). On the other hand, states which aggravate the stress increase the magnitude of additional changes, which then tend to drive the organism more strongly away from these states (Fig. 2a). This creates a statistical bias towards establishment of stress-reducing (i.e. beneficial) changes. This bias can be viewed as an extension of Le Chatelier’s principle [91], from chemical systems to biological self-organization. In this framework, stress provides not only a drive for variation but also a means for relaxation.
Since stress can be induced at (and have an impact on) every organizational level (e.g. cellular, tissue/organ and the whole organism level), stress-reductions can occur in multiple levels, enabling de novo emergence of coordination between adaptive improvisations in different cells, tissues and organs. As a basic principle of organization, it also applies to higher levels of organization (e.g. populations of individuals), but our main concern here is the organization within an individual.
Notably, the level of stress cannot be reduced to zero, because the over-abundance of changes which increase stress becomes stronger with decreased stress (Fig. 2a). At sufficiently low levels of stress, this over-abundance outweighs the effect of drive reduction, and random changes then lead to increase of stress. The most expected levels of stress therefore reside in an intermediate regime (indicated by the distribution function in Fig. 2a) in which the drive-reduction effect is roughly balanced by the entropic effect of higher abundance of stress-increasing states.
Stress alleviation by drive-reduction is further enhanced and stabilized when beneficial changes are also capable of reinforcing themselves by positive feedback mechanisms (Fig. 2b). An obvious example is provided by auto-regulatory processes in which beneficial changes are conferred by factors that are also capable of promoting their own production. Due to limited resources available to the organism, enhancement of these auto-regulatory changes can repress other (not necessarily beneficial) alternatives, thus leading to further selection and stabilization of the beneficial changes. This can expedite the process of coping with the stress, stabilize beneficial outcomes, and reduce the probability of more harmful changes. Stabilization by auto-regulation may also contribute to the persistence of the adaptive changes even when the inducing conditions revert, thus providing potential for maintaining these changes as ‘memorized’ adaptations that are immediately available upon re-introduction of the stressful conditions. Moreover, inter-individual exchange of beneficial auto-regulatory factors can reduce the level of exploration performed by each individual, thereby decreasing the potential for harmful changes in these individuals.
Stress can be reduced either by pre-evolved response programs or via the much slower process of adaptive improvisation. The relative contribution of each process is determined by the efficacy of the response, without any need for a computational-like ‘decision making’ processes. If the stress is ‘familiar‘ (e.g. heat shock, starvation, drought, etc.) and sufficiently moderate, it is quickly recognized by existing mechanisms and invokes efficient pre-evolved responses which rapidly alleviate the stress. This lowers the drive for random changes, thereby reducing the extent of adaptive improvisation. However, when the stress is novel or too strong, its alleviation by pre-evolved responses is compromised and the remaining stress increases the extent of adaptive improvisation.
A purely random variation has two opposing influences: It increases the probability of finding a solution to a new problem (adaptation), but also the likelihood of encountering deleterious changes. The adaptive potential of random variation is therefore expected to depend on the ratio between the probabilities of beneficial and deleterious outcomes of a given change:
To enable beneficial use of random variation, this ratio must be substantially higher than one. This requires a strong inverse correlation between the flexibility of traits and their deleterious potential (Fig. 1). The inverse correlation is an intrinsic outcome of natural selection which lowers the flexibility in directions that have been previously deemed deleterious. Due to this bias, random changes occur primarily in less deleterious traits, thus reducing the detrimental effect of the changes. The probability for beneficial (adaptive) random changes, on the other hand, is not preferentially dependent on the ability to vary along the more restricted directions. Consequently, the restriction of variation in critical traits suppresses deleterious changes much more than it prevents beneficial changes. The stronger the differential suppression of deleterious potential, the larger the bias towards less harmful random changes. A sufficiently strong bias reduces the detrimental effects enough to enable stress reduction (i.e. adaptation) by newly-forming random variation. However, it is important to realize that this bias towards beneficial random variation cannot guarantee adaptive outcome. No matter how large the bias is, it does not necessarily lead to adaptation by improvisation because the probability of non-beneficial changes cannot be reduced to zero. In other words, adaptive improvisation is a likely outcome, but not a guaranteed outcome.
The ability to restrict random variation has been suggested to occur even in genomic space [58]. The usefulness of this restriction depends, however, on the ability to differentially suppress changes in loci with high deleterious potential. Achieving this for genetic changes requires functional separation of the genome into readily changing elements with relatively low deleterious potential and rarely changing elements with high deleterious potential. Random mutagenesis in the latter are not beneficial to the individual, but if kept at sufficiently low rates, it may still be beneficial at the population level.
Below we propose specific examples of potential mechanistic implementations in both unicellular and multicellular organisms. While these mechanisms may have a significant contribution to adaptation, they are not the only mechanisms which can support adaptive improvisation. We therefore include these mechanisms only as a way of demonstrating the feasibility and generality of the hypothesis and to provide clear (but non-exclusive) examples of variations which satisfy the underlying assumptions of adaptive improvisation.