Jared Edward Reser PhD with GPT 5.6 Sol
Abstract
Evolutionary history contains many cases in which traits disappear, remain unexpressed for extended periods, or reappear in related forms. These patterns raise the possibility that genomes and developmental systems preserve more than presently expressed adaptations. They may also retain latent capacities that make previously useful phenotypes, or phenotypes assembled from previously useful components, unusually easy to recover. Several established concepts address parts of this possibility, including standing genetic variation, cryptic genetic variation, canalization, evolutionary capacitance, phenotypic plasticity, bet-hedging, developmental bias, facilitated variation, latent phenotypes, developmental memory, and the evolution of evolvability. Theory and experiment also show that natural selection can act indirectly on mechanisms that determine the production of future variation.
The stronger hypothesis considered here is that recurrent environmental change can favor genomic and developmental architectures that preserve access to historically adaptive regions of phenotype space. This is termed the historically structured latent repertoire hypothesis, a refinement of the more general selected latent repertoire hypothesis. The measurable property at issue is called phenotypic recoverability, defined as the probability, speed, or developmental accessibility with which a lineage can restore a previously expressed or closely related phenotype. When lineages possessing greater recoverability leave more descendants following environmental change, the process can be characterized as second-order lineage selection for phenotypic recoverability. At the species level, it is closely related to the established concept of species selection on variability.
Existing evidence supports most of the component processes required by this hypothesis, but it does not yet establish that animal lineages generally preserve unused adaptations specifically for future use. The central empirical question is whether latent variation is merely abundant, or whether it is historically organized so that previously adaptive phenotypes remain disproportionately accessible after controlling for pleiotropy, present function, neutral persistence, developmental complexity, and environmental recurrence.
Keywords: evolvability, cryptic genetic variation, developmental memory, facilitated variation, latent phenotype, lineage selection, phenotypic recoverability, species selection, evolutionary reversal
1. The problem of latent evolutionary capacity
Natural selection is usually described as a process acting on currently existing heritable differences. Some organisms survive or reproduce more successfully than others, causing the variants they carry to become more common. This formulation is correct, but it can obscure a second target of evolution: the developmental and genetic machinery that determines which variants will become available in later generations.
A lineage does not inherit only a collection of presently expressed traits. It also inherits a genotype-to-phenotype map, a developmental organization, a pattern of genetic covariation, a repertoire of reaction norms, and a distribution of mutational possibilities. These inherited properties affect which phenotypes can be produced after environmental change, recombination, mutation, hybridization, or developmental perturbation. Evolvability theory concerns this capacity to generate heritable phenotypic variation, especially variation capable of supporting adaptation.
The motivating observation is that traits sometimes appear to disappear and later return. Light pigmentation provides an intuitive example. A darkly pigmented population may continue to carry developmental mechanisms capable of producing light hair, fur, or natal coats. If light pigmentation later reappears, it can look as though an old capacity was kept in reserve.
Phenotypic resemblance, however, is not sufficient evidence for such preservation. Common blond hair in northern Europeans is associated with regulatory variation near KITLG, whereas blond hair in Solomon Islanders is strongly associated with a coding variant in TYRP1. Similar pigmentation therefore arose through at least partly distinct genetic routes. Distinctive natal coats in primates also have plausible current functions, including associations with social conditions such as infanticide risk. These cases motivate the hypothesis, but they do not by themselves demonstrate reactivation of an ancestral pigmentation program.
The broader question is not whether every recurring trait is an atavism. It is whether selection can shape biological systems so that some presently unexpressed phenotypes remain easier to produce than other equally complex possibilities. If the accessible possibilities reflect a lineage’s previous adaptive history, then evolutionary history may be stored not as a literal archive of finished traits, but as structure in the developmental search space.
2. The existing conceptual landscape
No single established term covers the entire hypothesis. Several literatures address different parts of it.
2.1 Evolvability and variability
Evolvability is the capacity of a biological system to generate heritable phenotypic variation that can support evolutionary change. It can be influenced by mutation rates, recombination, modularity, developmental architecture, robustness, pleiotropy, plasticity, gene regulation, population structure, and the topology of genotype networks. Evolvability is not simply the number of variants currently observed. It also concerns the phenotypes that a system could produce under genetic or environmental perturbation.
The related term variability refers to the propensity or capacity to vary, as distinct from realized variation among individuals. A population may presently display little variation while possessing a developmental architecture capable of producing many phenotypes under different circumstances.
This distinction closely matches the intuition of maintaining “variation in variability.” Selection can influence not only mean trait values, but also the amount, direction, covariance, and conditional expression of future variation.
2.2 Standing genetic variation
Adaptation often uses alleles that already exist in a population. This is known as adaptation from standing genetic variation. The relevant variants may currently be rare, mildly disadvantageous, recessive, locally favored, or repeatedly introduced through gene flow.
Threespine sticklebacks provide a classic example. Low-armor alleles around the EDA locus occur at low frequencies in marine populations and have repeatedly increased when marine fish colonize freshwater. The low-plated haplotype is ancient and provides a molecular basis for rapid, parallel phenotypic change. The marine population therefore contains variants that can be rapidly redeployed in an ecological setting where they become advantageous.
Standing variation can function as an evolutionary reserve without having evolved for that purpose. Rare alleles can persist because of migration, balancing selection, recessivity, weak costs, spatial heterogeneity, or drift. Demonstrating that variation exists is therefore easier than demonstrating that it has been selectively maintained because of its future adaptive value.
2.3 Cryptic genetic variation
Cryptic genetic variation, or CGV, consists of heritable differences with little or no phenotypic effect under normal conditions that become consequential in unusual environments or genetic backgrounds. It can accumulate while development buffers its effects, then become exposed by stress, mutation, hybridization, or changes in regulatory context.
CGV is one of the closest established concepts to a genomic reserve. A population can appear phenotypically uniform while carrying differences that become selectable after conditions change. Yet CGV need not be adaptively stored. Some of it may be neutral, some deleterious, and some useful only by coincidence. The stronger question is whether selection can influence the structure of CGV so that it is biased toward historically useful directions.
2.4 Canalization, robustness, and evolutionary capacitance
Canalization describes the production of a stable phenotype despite environmental and genetic perturbation. Robust development can conceal genetic differences, allowing cryptic variants to accumulate. If buffering later weakens, a process sometimes called decanalization, those variants may become phenotypically visible.
An evolutionary capacitor buffers variation under ordinary conditions and releases it under particular perturbations. Hsp90 is the classic example. Reduced Hsp90 function in Drosophila exposes background-dependent morphological variation affecting many structures. Later work has continued to examine whether such buffered variation can contribute to adaptation.
This creates a relationship between robustness and evolvability that initially seems paradoxical. A robust system resists phenotypic change in the present, yet that same resistance can permit hidden genetic diversity to accumulate. Once exposed, the stored diversity may accelerate evolutionary change.
2.5 Phenotypic plasticity, bet-hedging, and genetic accommodation
Phenotypic plasticity allows a genotype to produce different phenotypes in different environments. Plasticity does not wait for a new mutation. It uses environmental information to select among pre-existing developmental outcomes.
Plasticity can also alter later evolvability. Models show that the evolution of plasticity can increase mutational and genetic variance along the dimensions in which the organism has become developmentally flexible. A plastic response may subsequently undergo genetic accommodation, in which selection changes its magnitude, threshold, reliability, or regulation. In some cases, the induced phenotype becomes constitutively expressed through genetic assimilation.
When environments are unpredictable and no reliable cue is available, selection can favor bet-hedging, in which genetically similar organisms produce multiple phenotypic states. Experimental bacterial populations have evolved stochastic switching under fluctuating conditions, demonstrating that selection can favor the production of diversity itself rather than one fixed phenotype.
Plasticity and bet-hedging are strong examples of selection maintaining phenotypes that are not useful in every generation. Their continued existence is explained by environmental recurrence. A phenotype unexpressed for many generations can still contribute to long-term fitness if the conditions requiring it return often enough.
2.6 Facilitated variation and developmental bias
The theory of facilitated variation proposes that conserved, modular, and adaptable core processes make viable phenotypic change easier to generate. Evolution often does not construct a complex trait from entirely new components. It changes when, where, and how existing components are deployed. Weak regulatory linkage, modular organization, and exploratory developmental processes allow relatively modest regulatory changes to produce coordinated phenotypic effects.
The related concept of developmental bias emphasizes that genetic and environmental perturbations do not generate phenotypes uniformly. Development makes some directions of change much more accessible than others. Developmental bias is not necessarily an alternative to natural selection. The bias itself can evolve under selection and then influence later evolutionary trajectories. Empirical work has found that developmental bias can predict macroevolutionary patterns over long intervals.
These ideas replace the image of evolution searching all conceivable phenotypes at random. Evolution searches through a structured neighborhood created by the inherited developmental system.
2.7 Latent phenotypes and latent repertoires
The term latent phenotype already exists in the literature. A latent phenotype is an unexpressed phenotype that becomes accessible through genetic or regulatory change while the system retains its current function. Payne and Wagner examined millions of model gene-regulatory circuits and found that latent expression phenotypes were widespread. They used latent repertoire for the set of such accessible but currently unexpressed outputs. Robust regulatory systems often possessed larger repertoires.
The phrase “latent repertoire” is therefore not new. The proposed extension concerns why the repertoire has the structure it does. Is it simply a byproduct of robustness and network organization, or can selection preserve and organize it according to the lineage’s past selective environments?
2.8 Developmental memory and evolutionary generalization
Models of evolving regulatory networks have shown that developmental organization can acquire a form of developmental memory. Networks exposed to multiple selected phenotypes can evolve distributed interactions that reconstruct those phenotypes from incomplete starting conditions and produce novel combinations of previously selected features. The analogy is to associative memory in neural networks.
Other models show that evolution can “learn” regularities shared across past environments. When environmental variation is systematically structured, selection can favor developmental architectures that generalize to related environments not previously encountered. Conditions analogous to regularization in machine learning can prevent evolutionary overfitting to one recent environment.
This theoretical work is exceptionally close to the present hypothesis. It shows how apparent evolutionary foresight can emerge without foresight. A lineage does not predict a particular future. Its developmental organization internalizes recurrent statistical relationships from past selection, making related future adaptations easier to find.
3. From latent variation to a historically structured repertoire
The central proposal is that a developmental system’s latent repertoire may sometimes be historically structured.
A large latent repertoire alone does not imply historical memory. A regulatory network could produce many unexpressed phenotypes without any relationship between those phenotypes and the lineage’s adaptive past. The stronger hypothesis predicts that formerly useful phenotypes, or new phenotypes composed from formerly useful modules, remain unusually close in developmental and genetic space.
The historically structured latent repertoire hypothesis can be stated as follows:
Recurrently changing selection favors genomic and developmental architectures that preserve low-cost access to previously adaptive phenotypes and to new combinations of historically useful phenotypic components. As a result, latent variation is biased toward regions of phenotype space that reflect the lineage’s previous selective environments.
An earlier and broader label for the same idea is the selected latent repertoire hypothesis. The refined name adds the central prediction that the repertoire is not merely preserved, but organized by history.
Three claims are embedded in the hypothesis.
3.1 Historical structuring
The distribution of accessible latent phenotypes should contain information about previous selective environments. Past adaptations should influence which phenotypes remain easy to produce, even when those phenotypes are not currently expressed.
3.2 Selected retention
Mechanisms preserving access to latent phenotypes may sometimes be favored because they increase long-term descendant success. This is stronger than accidental persistence and stronger than preservation due solely to an unrelated present function.
3.3 Differential recoverability
A previously adaptive phenotype should, under specified conditions, be easier to restore than an equally complex phenotype with no history in the lineage. The difference could appear as fewer required mutations, smaller regulatory perturbations, faster selection response, greater developmental viability, or a higher probability of reconstruction.
The proposal is not that genomes contain complete dormant blueprints for every ancestral adaptation. A better model is a generative construction system. Genes, regulatory elements, signaling pathways, cell behaviors, anatomical modules, and developmental interactions are repeatedly reused. A phenotype may disappear while much of the machinery capable of reconstructing it remains active in other contexts.
The descriptive phrase evolutionary reserve can be used for this retained capacity, provided that “reserve” is understood as accessibility rather than storage of a finished object.
4. Phenotypic recoverability
The property at the center of the hypothesis is termed phenotypic recoverability.
Phenotypic recoverability is the capacity of a genotype, population, species, or lineage to regain a previously expressed phenotype, or a functionally related phenotype, after environmental or genetic change. It can be measured in several ways:
R_i(P,k \mid C) = \Pr(\text{lineage } i \text{ reaches phenotype } P \text{ within } k \text{ generations after challenge } C)
Here, R_i is the recoverability of phenotype P in lineage i, k is the time allowed for recovery, and C specifies the environmental or selective challenge.
Recoverability can also be represented as a distance:
D_i(P) = \text{minimum genetic, regulatory, developmental, or environmental change required to produce } P
High recoverability corresponds to a high probability of restoration, a short waiting time, or a small distance to the phenotype.
This definition separates recoverability from generic evolvability. A lineage may be highly capable of producing novelty but poor at returning to ancestral phenotypes. Another lineage may have limited overall variability but retain a short regulatory path to a particular previously useful state.
Recoverability can reside at different biological levels:
- Individual recoverability, where each organism inherits an integrated but suppressed developmental program.
- Population recoverability, where different individuals carry complementary variants or alternative adaptive alleles.
- Species recoverability, where geographically separated populations collectively preserve a broader repertoire.
- Clade recoverability, where conserved developmental organization makes particular phenotypic regions repeatedly accessible across descendant species.
These distinctions matter because the level at which the reserve is stored need not be the same as the level at which its consequences become visible.
5. Candidate mechanisms of retained recoverability
Several known mechanisms could produce the proposed pattern.
5.1 Regulatory suppression rather than structural deletion
Traits may disappear because of changes in timing, location, threshold, or intensity of gene expression. If downstream developmental machinery remains functional, reversal of an upstream regulatory change may restore substantial parts of the phenotype.
Regulatory suppression produces shallow trait loss. Deletion or degeneration of structural genes produces deeper loss. The hypothesis predicts that recurrently useful traits should disproportionately undergo shallow loss.
5.2 Pleiotropic maintenance
Genes formerly involved in one adaptation may remain under purifying selection because they serve other functions. This can preserve components of an old developmental pathway after one of its outputs disappears.
Pleiotropy therefore offers a non-foresighted mechanism for long-term recoverability. Selection maintains the components for current reasons, while their continued existence preserves access to additional phenotypes.
5.3 Modularity and co-option
Modular systems allow components to be altered or recombined without disrupting the entire organism. A module retained in one tissue can later be recruited into another. Many apparent reversals may therefore involve co-option of active machinery rather than resurrection of a completely dormant program.
5.4 Cryptic variation protected by robustness
Canalized development can accumulate conditionally silent alleles. Environmental stress, hybridization, or regulatory change may expose them. If previous selection has influenced which variants are tolerated within the robust network, the resulting CGV may be historically biased.
5.5 Standing variation and balancing processes
Alternative alleles may persist through spatially varying selection, temporal variation, introgression, migration-selection balance, dominance, or frequency-dependent selection. When conditions change, adaptation can proceed through rapid reweighting of existing variants.
5.6 Distributed species-level storage
Different populations may lose different components of a complex capacity. Hybridization or renewed gene flow can then reconstruct a functional phenotype. In such cases, the species collectively retains more recoverability than any one population.
5.7 Recurrent environmental activation
A latent program may be expressed too rarely for casual observation but often enough for selection to maintain it. Even very infrequent activation can preserve a pathway if the episodes are sufficiently consequential.
5.8 Selection on developmental architecture
Selection can change modularity, mutational sensitivity, regulatory connectivity, mutation localization, switching rates, and the covariance structure of phenotypic variation. The machinery producing future variants can itself become an object of evolution.
6. Illustrative biological cases
The following cases do not all provide evidence for selection specifically preserving dormant ancestral adaptations. They illustrate different components of the hypothesis.
6.1 Inducible defenses in Daphnia
Juvenile Daphnia pulex exposed to chemical cues from Chaoborus predators can develop neckteeth and associated defensive morphology. These structures reduce predation risk but have demographic or life-history costs. The defense is therefore expressed conditionally rather than constitutively.
This is a clear case of an adaptive program maintained in an off state. Its value does not lie in the present phenotype of an unexposed individual. Its value lies in the ability to produce a different phenotype when a recurrent environmental cue appears.
6.2 Locust phase polyphenism
Locusts exhibit density-dependent phase polyphenism. Low-density populations occupy a relatively solitary phase, whereas crowding can trigger a coordinated gregarious phase involving behavioral, physiological, morphological, and coloration changes. This is not a single latent feature but an alternative whole-organism strategy.
The locust developmental system preserves both strategies because ecological conditions repeatedly alternate. The example demonstrates that selection can maintain elaborate suites of phenotypes that are unexpressed during long intervals.
6.3 Conditional fat synthesis in parasitoid wasps
The parasitoid wasp Leptopilina heterotoma can switch lipogenesis on when developing in fat-poor hosts and off when developing in fat-rich hosts. What had appeared to be loss and re-evolution of fat synthesis can therefore reflect environmentally controlled plasticity. Models associated with this work indicated that rarely activated switches can remain functional over many generations.
This case is particularly relevant because it shows how an apparently absent metabolic capacity can remain intact behind a regulatory switch. It also cautions against inferring evolutionary loss solely from expression under one environmental condition.
6.4 Stickleback armor
Marine sticklebacks carry ancient low-armor alleles that can rapidly increase after freshwater colonization. Here the reserve is not a dormant developmental program carried identically by every individual. It is standing variation maintained at the population or metapopulation level.
The case shows that a species can contain multiple ecological solutions simultaneously, even when one is rare in the present habitat.
6.5 Snowshoe hare winter coloration
Most snowshoe hares become white during winter, but some populations remain brown. Winter coat color is associated with regulatory variation around Agouti, and a winter-brown allele appears to have entered snowshoe hares through introgression from black-tailed jackrabbits. The allele provides an existing adaptive option in areas with little persistent snow.
This is another population-level reserve. It also illustrates how hybridization can replenish latent variation and transfer previously tested developmental solutions across species boundaries.
6.6 Reconstruction of vision in cavefish hybrids
Different blind populations of Astyanax mexicanus lost vision through partly different genetic changes. Crosses between some cave populations produce offspring with improved eyes and measurable visual responses because functional alleles from one population complement defects in the other.
No single blind population necessarily carries a complete dormant visual system. The capacity is distributed. At the species level, enough intact components remain to reconstruct partial vision through recombination.
This provides an important extension of the reserve concept. Latent capacity can be a collective property of a species rather than a property of each organism.
6.7 Tooth-development potential in chickens
Modern birds lack teeth, yet the talpid2 chicken mutant initiates archosaurian tooth-like structures. Activation of beta-catenin signaling in ordinary chicken embryos can reproduce early events in this process. The work suggests that avian tooth loss involved changes in developmental signaling and tissue relationships while some odontogenic potential remained.
This does not demonstrate an intact dormant program for complete functional teeth. It shows that a deeply ancestral developmental output can remain surprisingly close in developmental space because the underlying signaling systems and competent tissues continue to participate in other processes.
6.8 Lower-jaw teeth in Gastrotheca guentheri
Most frogs lack mandibular teeth, and phylogenetic reconstruction indicates that lower-jaw teeth were lost in the common ancestry of frogs more than 200 million years ago. Gastrotheca guentheri possesses true teeth on the lower jaw. Morphological and developmental studies support an evolutionary regain, although interpretation of such deep reversals always depends on phylogenetic and developmental evidence.
The general tooth-development system remained active in the upper jaw. The lower-jaw phenotype could therefore have reappeared through redeployment of retained machinery. The parts were conserved even though one anatomical use had vanished.
6.9 Reactivation of mammalian regeneration
A 2025 study compared regenerative and nonregenerative mammalian species and identified deficient activation of Aldh1a2, involved in retinoic acid synthesis, as a limiting factor in mouse ear-pinna regeneration. Activating the pathway or introducing a rabbit regulatory enhancer improved regeneration in transgenic mice.
This is one of the clearest experimental demonstrations that a complex biological capacity can remain partly available after evolutionary suppression. A regulatory intervention restored access to downstream machinery that had not been completely lost.
6.10 Light hair and natal coloration as tests rather than proofs
Light hair remains a useful conceptual example, but its interpretation requires genetic detail. The independent involvement of KITLG in European blond hair and TYRP1 in Solomon Islander blond hair shows that a similar phenotype can arise through different pathways. Such convergence would not support preservation of one ancestral blond-hair program.
Primate natal coats present a related challenge. A light or contrasting infant coat may be produced by age-specific pigmentation regulation, but comparative evidence points to ongoing social functions in at least some lineages. A trait expressed during infancy is not currently unused merely because it disappears in adults.
These examples clarify what evidence is needed. A recurring appearance is only the beginning. The hypothesis requires reconstruction of the causal developmental path.
7. Selection of evolvability and selection for evolvability
A central distinction is the difference between selection of an evolvable system and selection for evolvability.
Selection of evolvability occurs when lineages happen to differ in their capacity to evolve and the more evolvable lineages fare better after environmental change. Their evolvability may have originated as a byproduct of robustness, modularity, large population size, present-day plasticity, or some other property.
Selection for evolvability occurs when the machinery producing or preserving adaptive variation increases because its contribution to future adaptation causes the lineages carrying it to leave more descendants.
The distinction is difficult to establish empirically because the same mechanism may have immediate and delayed benefits. Hsp90, for example, has essential current physiological functions. Any evolutionary capacitance it provides may initially be a byproduct of those functions.
Experimental microbial evolution provides stronger cases. Woods and colleagues showed second-order selection for greater evolutionary potential among competing Escherichia coli lineages. In 2025, Barnett, Meister, and Rainey reported the evolution of localized hypermutation in bacteria exposed to recurring transitions between phenotypic states. The evolved architecture increased mutation specifically in a region where reversible variation was useful.
These experiments establish that selection can alter the generation of future variants. They do not yet show that complex animal lineages routinely preserve ancestral phenotypes, but they remove the theoretical objection that evolvability cannot itself be selected.
8. The level of selection
The proposed process is not wholly outside individual selection and group selection. It belongs within a multilevel framework, but the relevant unit is often extended through time.
8.1 Organismal and gene-level selection
Suppose every organism carrying an allele also carries a latent program that occasionally becomes useful. If the program is activated within the organism’s lifetime, or if its benefits reliably return to genetically associated descendants, ordinary organismal or gene-level selection may be sufficient.
A gene-centered account can therefore explain some selection for evolvability. A modifier that increases the production of beneficial variants can spread if it remains associated with those variants.
The difficulty is genetic association. In sexual populations, recombination can separate an evolvability-enhancing modifier from the adaptive alleles it helped produce. This makes indirect selection weaker than in asexual lineages. Integrated developmental architectures, low recombination, population structure, or recurrent activation can preserve the association.
8.2 Group selection
If populations differ in the diversity they maintain, and populations with greater diversity survive disturbances more often, the process may be represented as group or deme selection. This is most appropriate when contemporaneous populations reproduce, split, or go extinct at different rates.
8.3 Lineage selection
A lineage is genealogically extended through generations. Lineage selection occurs when heritable lineages differ in a property that affects their long-term persistence or proliferation.
This level is especially appropriate for evolvability because the causal sequence extends through time:
\text{inherited architecture} \rightarrow \text{later production or recovery of adaptive phenotypes} \rightarrow \text{greater descendant representation}
The descendants do not act backward in time. At the initial moment, lineages differ in the probability distributions of their possible futures. After environmental change, lineages with higher recoverability are more likely to remain represented.
8.4 Species selection on variability
Lloyd and Gould introduced the established phrase species selection on variability to describe cases in which within-species variability affects the long-term persistence or proliferation of species. Genetic variability is physically distributed among organisms, but its amount can function as an aggregate species-level character.
If species with richer or better organized latent repertoires are less likely to go extinct or more likely to produce descendant species, the present hypothesis becomes a form of species selection on latent variability.
8.5 Clade selection for evolvability
At still larger scales, clades may differ in developmental modularity, ecological flexibility, or accessible phenotypic variation. Jablonski has argued that evolvability is appropriately studied from a multilevel macroevolutionary perspective by comparing clades exposed to similar opportunities, such as mass extinctions or entry into new adaptive zones. Holstad and colleagues found that measured evolvability predicts phenotypic divergence among populations and, more weakly, among species across contemporary and fossil datasets.
If clade-level differences are heritable and affect differential persistence or diversification, clade selection for evolvability becomes an appropriate description.
9. Proposed terminology
The existing literature already provides the broad category, selection for evolvability, and the process category, second-order selection. It also provides latent phenotype, latent repertoire, developmental memory, and species selection on variability.
The working terms proposed here are narrower.
Historically structured latent repertoire hypothesis
The hypothesis that selection in recurrently changing environments preserves and organizes latent phenotypic possibilities according to the lineage’s adaptive history.
Selected latent repertoire hypothesis
A shorter and somewhat broader name for the proposition that latent phenotype-generating capacity can itself be selectively maintained.
Phenotypic recoverability
The probability, speed, or developmental accessibility with which a lineage can restore a previously expressed or functionally related phenotype.
Lineage selection for phenotypic recoverability
Differential persistence or proliferation of heritable lineages caused by differences in their ability to retain and re-express presently absent phenotypes.
Second-order lineage selection for phenotypic recoverability
The most precise name for the proposed process. “Second-order” identifies selection on the production of future adaptive possibilities. “Lineage” identifies the temporally extended entity through which the benefit is realized. “Phenotypic recoverability” identifies the specific evolvability property being favored.
Lineage selection for latent evolvability
A broader alternative that includes capacities for both ancestral recovery and related innovation.
These labels are proposed as working terminology rather than as a claim that their conceptual content has no precedent. Their value would lie in separating historically informed recoverability from generic evolvability.
10. A mathematical characterization
Let R_i denote the recoverability of lineage i, and let W_i(T) denote its descendant representation after T generations.
Selection for recoverability occurs when:
\operatorname{Cov}_{i}\left(R_i,W_i(T)\right) > 0
The defining case would be one in which recoverability has little or no immediate organismal benefit:
\operatorname{Cov}_{\text{individuals}} \left(R_i,w_i(1)\right) \leq 0
while still producing a positive long-term covariance among lineages:
\operatorname{Cov}_{\text{lineages}} \left(R_i,W_i(T)\right) > 0
The condition does not imply that future events cause present preservation. It states that lineages with different inherited architectures have different expected long-term outcomes.
A simple heuristic for the strength of selection on a latent capacity is:
S_R \propto qBL – C_m – C_e
where:
- q is the frequency with which relevant environmental conditions recur,
- B is the benefit of rapid phenotype recovery,
- L is the strength of genealogical or genetic association between the recoverability mechanism and the descendants it benefits,
- C_m is the maintenance cost of the latent machinery,
- C_e is the cost of erroneous or inappropriate activation.
Recoverability is most likely to evolve when environmental challenges recur, the advantage of rapid adaptation is large, maintenance costs are low, activation is accurately regulated, and the recoverability architecture remains associated with the successful descendants it generates.
The strongest evidence for higher-level selection would occur when recoverability is costly to current individuals but beneficial to the persistence of populations, species, or clades.
11. Predictions of the hypothesis
The historically structured latent repertoire hypothesis generates predictions beyond the general claim that organisms possess hidden variation.
11.1 Recurrently useful traits should undergo shallower loss
Traits associated with recurring environmental conditions should more often disappear through regulatory suppression than through deletion of core developmental machinery. Their recoverability should decay more slowly than that of traits linked to environments that have permanently vanished.
11.2 Past adaptations should remain unusually close in developmental space
After controlling for trait complexity, pleiotropy, and current gene function, a previously adaptive phenotype should require fewer regulatory or genetic changes to recover than a phenotype of comparable complexity that was never expressed in the lineage.
11.3 Cryptic variation should be historically biased
Cryptic variation should cluster around previously occupied adaptive regions rather than being distributed without relation to selective history. Perturbations should reveal ancestral or ancestrally related phenotypic combinations more often than appropriate null models predict.
11.4 Structured fluctuation should produce stronger generalization than random fluctuation
Lineages evolved across environments sharing repeated structural regularities should adapt more rapidly to a new environment sharing those regularities. Randomly changing environments may favor generic robustness or bet-hedging, but should not produce the same historically organized repertoire.
11.5 Repeatedly reversible traits should retain regulatory and developmental infrastructure
Comparative genomics should find unusually conserved enhancers, signaling relationships, or developmental competencies associated with traits that have repeatedly disappeared and reappeared.
11.6 Recoverability should predict persistence after environmental upheaval
Among related populations or clades exposed to similar change, those with greater measured recoverability should show lower extinction risk, faster ecological expansion, or greater diversification.
11.7 Deleting apparently redundant components should reduce future adaptability
Removing a presently dispensable component of a developmental system may have little effect under current conditions but reduce adaptation to historically recurrent challenges. This would provide experimental evidence that apparent redundancy contributes to an evolutionary reserve.
12. Experimental and comparative tests
A direct experimental test could evolve replicate populations under four environmental regimes:
- a constant environment,
- two predictably alternating environments,
- several environments sharing hidden structural regularities,
- several unrelated environments changing randomly.
After prolonged evolution, every lineage would be challenged with new environments. Some challenges would share components with the historical environments, while others would be unrelated.
The hypothesis predicts that structured environmental variation will produce developmental or regulatory systems that recover familiar phenotype components and generalize to related new combinations. Constant environments should favor specialization and permit latent capacities to decay. Random environments may favor broad tolerance or stochastic switching, but not the same pattern of historically specific accessibility.
The experiment could be conducted in microbes, rapidly reproducing animals, organoids, synthetic gene networks, or digital developmental systems. Genome sequencing, regulatory perturbation, and lineage tracking could determine whether adaptation involves standing variation, cryptic variants, localized mutability, modular regulatory organization, or reversible switches.
Comparative tests would require reconstructing trait histories across clades. Researchers could estimate how often traits are lost and regained, whether loss is regulatory or structural, how rapidly associated genes decay, and whether recurrence correlates with environmental periodicity. Closely related clades exposed to similar opportunities would provide the most informative contrasts.
A particularly strong design would compare the accessibility of two phenotypes with similar developmental complexity, one historically expressed and one counterfactual. Greater accessibility of the historical phenotype would support developmental memory. Evidence that the accessibility is maintained despite a measurable cost would support selection for recoverability.
13. Alternative explanations and limits
The principal difficulty is distinguishing selected preservation from incidental persistence.
A dormant capacity may remain because its components have other current functions. It may be inexpensive to retain. Too little time may have elapsed for degeneration. The phenotype may still be expressed in an overlooked tissue, developmental stage, sex, or environmental context. Gene flow may repeatedly restore lost alleles. The apparent recurrence may be convergent evolution rather than reactivation.
These alternatives are not minor complications. They are often the most likely explanations.
Dollo’s law and related work on evolutionary irreversibility provide an important boundary condition. Once a pathway is released from selection, mutations can degrade genes, enhancers, developmental interactions, and anatomical contexts. Reversal becomes progressively less likely as these dependencies disappear or become reorganized. Evolutionary recovery is therefore expected to be time-sensitive and partial rather than unlimited.
A useful distinction is between three depths of loss:
- Shallow loss, where expression is suppressed but most machinery remains.
- Intermediate loss, where components remain through pleiotropy or reuse, but the original organization has partly degraded.
- Deep loss, where essential genes, regulatory elements, or developmental competencies have disappeared.
The hypothesis should apply most strongly to shallow and intermediate loss. It does not predict indefinite preservation of every ancestral adaptation.
Another limitation concerns the word “adaptive.” A latent repertoire may contain deleterious, neutral, and useful possibilities. The presence of many hidden phenotypes does not imply that the repertoire is optimized. Even developmental memory models can overfit past environments, preserve obsolete correlations, or constrain access to genuinely novel solutions.
The question is therefore comparative and statistical. Are historically useful phenotypes more recoverable than plausible alternatives, after controlling for ordinary conservation and developmental reuse?
14. Relation to gene-centered evolution
The proposed hypothesis does not require abandoning the selfish-gene perspective. Gene-level selection, organismal selection, lineage selection, and species selection are alternative descriptions of causal structure, not necessarily mutually exclusive theories.
A modifier allele that preserves a latent pathway can spread through gene-level selection if it remains associated with the later adaptations it enables. Individual-level benefits can maintain the same pathway for immediate reasons. Population and lineage-level consequences can arise from the collective distribution of those alleles.
The gene-centered account becomes less complete when the selected property is an aggregate feature, such as population-wide genetic diversity, distributed complementation, or an integrated developmental architecture whose consequences appear only over long timescales. In those cases, a multilevel description captures causal relationships that are difficult to express solely as immediate competition among individuals.
The most defensible position is therefore not that phenotypic recoverability constitutes a wholly new force of evolution. It is a proposed target of selection that may be represented at different levels depending on how the capacity is inherited and where differential persistence occurs.
15. The appearance of evolutionary foresight
Natural selection has no awareness of natural selection, and no future organism causes its ancestors to preserve a trait. Yet systems shaped by repeated environmental change can acquire behavior that resembles anticipation.
There are three routes to this appearance.
First, recurrent environments favor conditional programs. The organism responds to a cue because similar cues were informative in ancestral environments.
Second, bet-hedging favors diversified outcomes when future conditions are uncertain.
Third, developmental memory and evolutionary generalization allow regulatory architecture to internalize recurring relationships among selected phenotypes. The system becomes biased toward solutions that were useful in the past or that reuse the same structural regularities.
The resulting process can be described as functional foresight, but it is produced by historical filtering. Evolution does not know which phenotype will be useful. It can nevertheless construct a search space in which certain classes of useful phenotype are easier to find.
This may be the deepest implication of the hypothesis:
Natural selection can alter the machinery that supplies possibilities to later natural selection, causing past selection to influence the direction and accessibility of future evolutionary change.
16. The role of advanced scientific artificial intelligence
Testing the hypothesis across deep time will require integration of evidence that is currently divided among paleontology, comparative genomics, developmental biology, population genetics, ecology, and experimental evolution.
A sufficiently capable scientific AI could reconstruct latent genotype-to-phenotype maps using:
- fossil histories of trait loss and recurrence,
- ancestral genome and protein reconstruction,
- comparative enhancer and regulatory-element maps,
- developmental single-cell atlases,
- spatial gene-expression data,
- protein interaction networks,
- perturbation screens,
- population genomic variation,
- hybridization histories,
- and reconstructions of past environments.
The decisive advance would be the ability to estimate counterfactual accessibility. For a living lineage, the system could ask how many changes would be required to produce a previously expressed phenotype, an ancestrally related phenotype, or a phenotype never present in the lineage. It could then compare observed accessibility against models based on neutral persistence, pleiotropic maintenance, developmental constraint, recurrent direct selection, and selection for recoverability.
Artificial systems could also identify regulatory interventions that distinguish shallow from deep trait loss. A phenotype restored by one enhancer substitution has a different evolutionary status from a phenotype requiring reconstruction of dozens of degraded components.
Some historical information has been permanently erased and cannot be recovered with certainty. Even so, the hypothesis need not be tested through complete reconstruction. Comparative likelihoods, predicted signatures, and targeted experiments could establish whether selection for recoverability explains patterns that simpler models do not.
17. Conclusion
The idea that species retain unused adaptations for future descendants is too strong if interpreted as deliberate storage of complete phenotypic programs. Genomes are not museums, and natural selection does not preserve every obsolete feature. Unused pathways decay, similar traits can evolve independently, and many latent capacities persist because their components remain useful in other contexts.
A more defensible and scientifically productive claim is available.
Genomes and developmental systems inherit structured spaces of phenotypic possibility. Standing variation, cryptic genetic variation, robustness, plasticity, modularity, facilitated variation, developmental bias, and regulatory memory can all preserve access to phenotypes that are not currently expressed. Theory and experiment show that natural selection can modify these variation-generating systems and can favor greater evolutionary potential. Species and clades may also differ in variability, recoverability, and long-term persistence.
The proposed extension is that recurrent selection may sometimes preserve historically structured latent repertoires. Such repertoires would make previously adaptive phenotypes and related combinations easier to recover than arbitrary alternatives. The property can be called phenotypic recoverability, and its long-term selection can be described as second-order lineage selection for phenotypic recoverability.
The hypothesis is not yet established as a general principle. Its component mechanisms are well supported, its theoretical foundation is credible, and it yields discriminating empirical predictions. The central question is no longer whether organisms contain hidden variation. They clearly do. The deeper question is whether the organization of that hidden variation records a lineage’s adaptive history and whether natural selection has sometimes preserved that organization because of the evolutionary futures it makes possible.
Selected references
Barnett, M., Meister, L., and Rainey, P. B. 2025. Experimental evolution of evolvability. Science.
Beaumont, H. J. E., Gallie, J., Kost, C., Ferguson, G. C., and Rainey, P. B. 2009. Experimental evolution of bet hedging. Nature 462: 90-93.
Borowsky, R. 2008. Restoring sight in blind cavefish. Current Biology 18: R23-R24.
Colosimo, P. F., et al. 2005. Widespread parallel evolution in sticklebacks by repeated fixation of ectodysplasin alleles. Science 307: 1928-1933.
Gerhart, J., and Kirschner, M. 2007. The theory of facilitated variation. Proceedings of the National Academy of Sciences 104: 8582-8589.
Harris, M. P., et al. 2006. The development of archosaurian first-generation teeth in a chicken mutant. Current Biology 16: 371-377.
Holstad, A., et al. 2024. Evolvability predicts macroevolution under fluctuating selection. Science 384: 688-693.
Jablonski, D. 2022. Evolvability and macroevolution: overview and synthesis. Evolutionary Biology 49: 265-291.
Kouvaris, K., Clune, J., Kounios, L., Brede, M., and Watson, R. A. 2017. How evolution learns to generalise: using the principles of learning theory to understand the evolution of developmental organisation. PLoS Computational Biology 13: e1005358.
Lloyd, E. A., and Gould, S. J. 1993. Species selection on variability. Proceedings of the National Academy of Sciences 90: 595-599.
Marshall, C. R., Raff, E. C., and Raff, R. A. 1994. Dollo’s law and the death and resurrection of genes. Proceedings of the National Academy of Sciences 91: 12283-12287.
Paaby, A. B., and Rockman, M. V. 2014. Cryptic genetic variation: evolution’s hidden substrate. Nature Reviews Genetics 15: 247-258.
Parter, M., Kashtan, N., and Alon, U. 2008. Facilitated variation: how evolution learns from past environments to generalize to new environments. PLoS Computational Biology 4: e1000206.
Payne, J. L., and Wagner, A. 2014. Latent phenotypes pervade gene regulatory circuits. BMC Systems Biology 8: 64.
Payne, J. L., and Wagner, A. 2019. The causes of evolvability and their evolution. Nature Reviews Genetics 20: 24-38.
Rutherford, S. L., and Lindquist, S. 1998. Hsp90 as a capacitor for morphological evolution. Nature 396: 336-342.
Szilágyi, A., Szabó, P., Santos, M., and Szathmáry, E. 2020. Phenotypes to remember: evolutionary developmental memory capacity and robustness. PLoS Computational Biology 16: e1008425.
Uller, T., Moczek, A. P., Watson, R. A., Brakefield, P. M., and Laland, K. N. 2018. Developmental bias and evolution: a regulatory network perspective. Genetics 209: 949-966.
Watson, R. A., Wagner, G. P., Pavlicev, M., Weinreich, D. M., and Mills, R. 2014. The evolution of phenotypic correlations and developmental memory. Evolution 68: 1124-1138.
Woods, R. J., et al. 2011. Second-order selection for evolvability in a large Escherichia coli population. Science 331: 1433-1436.
Leave a Reply