1. Purpose and patterns

The general purpose of AEGIS is to simulate evolution of life history and aging under diverse environments, mortality structures, intrinsic constraints, tradeoffs/ genetic architectures, and population dynamics. The model’s suitability can be tested by its ability to reproduce various life history and aging patterns, including the evolution of age-dependent hazard rates (e.g., manifested as increasing intrinsic mortality with age), mortality plateaus, evolution of reproductive aging (manifested as decreasing or varying fertility with age), oscillating population sizes, evolution of the mutation rate [40], and convergent evolution of life history (Results).

2. Entities, state variables, and scales

The model contains two entities – environment and individual. There is one instance of the environment entity and many instances of the individual entity, constituting the population.

The central subject of interest is the evolution of individuals inhabiting a shared environment (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. State variables and scales of the two model entities (environment and individual), following the ODD protocol.

https://doi.org/10.1371/journal.pcbi.1014109.t001

3. Process overview and scheduling

This section outlines model processes, the order of their execution and the rationale for their inclusion.

The model runs in three phases: initialization, simulation and termination.

During the initialization phase, the population is created (details in section 5).

During the simulation phase the population evolves. This phase runs in steps whose number is defined by an input parameter, by default set to 1 million. During each step, a list of processes is executed: some, conceptually concerning the environment (e.g., growth of the predator population), indirectly affect individuals; others concern the individuals directly (e.g., individuals mating). Ultimately, all processes together impact the survival and/or reproduction of individuals in the population. The order of processes given default parameters is illustrated in Fig 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Experimental design framework in AEGIS.

AEGIS simulations are defined by three core components—parameterization, initialization, and termination—applied to a baseline simulation and one or more comparative scenarios. Differences in evolved life-history traits can then be causally attributed to changes in evolutionary conditions.

https://doi.org/10.1371/journal.pcbi.1014109.g001

Metadata about the simulation run is recorded during the termination phase.

1. Mortality

Individuals can die from different causes, modeled as one of the five processes listed below.

2. Reproduction

Reproduction is critical for keeping the population alive and evolving. The probability of reproducing is dependent on age, genetics, and availability of mates if reproduction is sexual. Each reproducing individual can produce at most one offspring during one reproductive event. Each individual can reproduce at every simulation step beyond the age corresponding to sexual maturity.

Non-customizable reproduction mechanics include random mating and no parental care.

Customizable reproduction mechanics include asexual vs. sexual reproduction (including assortment), recombination rate, germline mutation rate (parameterized or evolved), viviparity vs. oviparity (including incubation time), age at maturity, age at menopause.

Under viviparity, generations are overlapping. If oviparity is activated, there will be an egg life stage during which the developing individuals are invulnerable to all mortality sources. Under oviparity, generations can be overlapping or non-overlapping, depending on user input.

3. Aging

This process increments the age variable of each individual, resulting in a + 1 age at each discrete simulation step.

4. Environmental drift

As a default setting, environmental drift is deactivated, but it can be turned on. Conceptually, environmental drift simulates long-term environmental change such as climate change, resource depletion, pollution, etc.

The main use of environmental drift is to allow the population to keep evolving adaptively. When the environment does not change, the fitness landscape is static. After the initialization of each simulation, populations undergo an evolutionary phase, away from the initial conditions, in response to the initialization parameters, such as population size, reproduction modality, mutation rate, etc. Once populations approach a stable state, natural selection acts mostly to purify new detrimental mutations, since the environment is not programmed to change. However, if environmental drift is activated, the environment undergoes shifts, which might change accordingly the fitness landscape over time. With active environmental drift, populations keep evolving adaptively, following their fitness peak.

Order of processes

Under the default parameters, mortality processes are scheduled as illustrated in Fig 2, in the following order: intrinsic mortality, abiotic mortality, infection, predation, starvation, reproduction, aging, and environmental drift. The order of mortality processes can be customized by the user.

Although the order in which mortality sources are applied will not affect the total number of deaths, it can impact the death structure and thus affect demography.

For example, in a population of 1000 individuals, a first source with 40% mortality results in 400 deaths, and a second source with 80% mortality causes 480 deaths, as it acts on the remaining 600. The observed, population-level mortality rates are then 40% and 48% (rather than 40% and 80%). The order of these mortality events does not change the total number of deaths (880 in both cases), corresponding to a total mortality rate of 88%. However, it does alter the death structure (40% + 48% vs. 8% + 80%), which may affect the population demography if, for example, one source of mortality impacts all age groups equally, while the other disproportionately affects older individuals.

Order of individuals

The order in which individuals are simulated can impact life history outcomes, similar to how process computation order influences the effects of each process. In many individual-based models, outcomes for individuals are computed sequentially—survival, reproduction, and other processes are simulated for one individual at a time, moving through the population until all individuals have been processed. The order of individuals in this sequential approach matters, as earlier computations can influence the outcomes of individuals simulated later. For example, the first individual seeking a mate has access to a larger pool of potential mates, while the last may face limited or no mating opportunities, potentially reducing fertility. If the order is not randomized but based on traits like age, this can introduce bias into the simulation.

In contrast, AEGIS uses parallel computation, where processes are computed simultaneously for all individuals. This removes the impact of one individual’s life history outcomes on another’s within the same simulation step.

4. Design concepts

5. Experimental design: Parameterization, initialization and termination

Experiments in AEGIS follow a simple experimental design that specifies a baseline simulation and one or more comparative simulations, and that considers three components – parameterization, initialization and termination (Fig 1). After running comparative simulations, users can compare the evolved life history traits across these simulations and causally link their differences to variations in evolutionary conditions.

The first step is to appropriately parameterize simulations, i.e., adjust default parameter values, to emulate the evolutionary conditions of interest. Each evolutionary experiment requires at least two parameter sets – one representing the baseline conditions, and one or more capturing comparative evolutionary scenarios of interest. The built-in parameters and their explanations are listed in S1 File Supplementary Information.

The next step to consider is initialization. AEGIS does not require empirical life history traits as input; instead, it generates baseline populations de novo. While users can specify initial population-level life histories through adjustable parameters—useful when investigating evolutionary dynamics from a specific phenotypic starting point, such as assessing whether a non-aging population can maintain its non-aging phenotype—in most cases, a burn-in simulation is recommended. A burn-in simulation functions as the control condition against which experimental runs can be meaningfully compared.

Not using a burn-in simulation as a baseline for comparison may result in biased interpretations. For instance, if a non-aging survival phenotype is set as the initial condition, and multiple comparative scenarios are simulated where the user observes that the lifespan decreases at varying rates, it would be incorrect to attribute the observed reductions in lifespan solely to the varied parameter. This is because the interpretation of lifespan reduction is relative to and contingent on the high initial survival condition. In a population with a low initial survival, the opposite might be observed, i.e., population lifespan might increase. Thus, the same parameterization can lead to different interpretations depending on the choice of the comparative baseline. The burn-in simulation produces a population that has reached equilibrium based on the values of all parameters that will remain unchanged in subsequent experimental runs. This ensures that any evolved differences during experimental runs can be unambiguously attributed to interventions in parameter values of those same experimental runs. Analogously, at the genomic level, burn-in simulations help establish a mutation-selection-drift balance, which is critical when evolutionary dynamics are studied.

Finally, users need to decide on the simulation termination point, i.e., the number of steps for the simulation to run. The total number of simulation steps represents a tradeoff: fewer steps result in faster simulations and reduced storage use but may provide insufficient time to observe relevant evolutionary outcomes or achieve equilibrium. To determine if equilibrium has been reached, it is best to compare the life history traits of interest over time – if no significant changes are observed after a considerable period, the population has likely reached equilibrium (S1 File Supplementary Information).

6. Input data

Input data consists of the list of customized parameters and, optionally, a file containing a pre-evolved population. Input data can be entered via the graphical user interface, as a text file in YML format when AEGIS is used via a terminal, or as a dictionary when AEGIS is used via scripting.

7. Output data

Simulation outputs in AEGIS are organized into three main categories: demographic records, life-history traits, and variant frequencies.

Demographic tables include longitudinal records of population age structure, age-specific death distributions by cause, and birth distributions by parental age. From these, population-level life-history measures (e.g., median lifespan and life expectancy) can be derived, many of which are directly included in the output and visualized in the GUI. Demographic data are also amenable to further analysis using life history theory tools such as Leslie matrices.

In addition to population-level data, AEGIS records individual life-history trajectories, including birth time, age at death, and reproductive output. Importantly, it also tracks intrinsic survival and reproduction probabilities, which reflect biological potential independent of realized outcomes, allowing users to disentangle genetic from environmental contributions to life history.

Finally, AEGIS provides pseudogenomic data for each individual, enabling exploration of the genetic architecture underlying both individual- and population-level life histories. This resolution supports evolutionary analyses of trait dynamics shaped by selection and drift. With appropriate preprocessing, standard population genetic methods can be applied to investigate selective sweeps, linkage disequilibrium, and allele frequency change.

AEGIS generates emergent demographics and life history traits

First, we ask whether aging can evolve spontaneously starting from a non-aging population within the AEGIS framework – in particular, we study how mortality and fertility curves are shaped during the simulation. Since our aim here is to demonstrate capabilities of AEGIS, rather than to replicate life history evolution in a particular scenario or species, we ran following simulations using default parameters (S1 File Supplementary information: Default parameter values). Here, default parameters broadly model humans – e.g., simulated populations reproduce sexually after a maturation phase, they exhibit overlapping populations – with some simplifications such as random mating and no simulation of parental investment (S1 File Supplementary Information).

We computed age-specific mortality and fertility by dividing the number of births (Fig 3b) and deaths (Fig 3c) for each age class (Fig 3a) over the total number of births and deaths for a given simulation time interval. Compared to the initial non-aging phenotypes, characterized by age-flat mortality and fertility, evolved populations display age-increasing mortality (Fig 4b) and age-decreasing fertility (Fig 4c), indicating that somatic and reproductive aging can spontaneously evolve in AEGIS. We detect these age-specific patterns after 250 generations (S2d-e Fig; 5,000 simulation steps). Simulated populations reach an equilibrium by generation 20,000 (S2g-h Fig; 240,000 steps), after which they are maintained until generation 200,000 (Fig 3; 3 million steps) when we finally terminated them.

Notably, in the current experiment, mortality increases unevenly across different ages. Several species [42], including humans [43], have been documented to exhibit a late-life plateau in mortality rates, suggesting the intriguing possibility that aging may decelerate or even halt in later life. Within AEGIS, we observe a similar flattening of the evolved mortality curve in late life (Fig 4b), akin to the mortality plateau documented in extreme human survivors, indicating that AEGIS could be used to study factors shaping the evolution of mortality deceleration.

Mortality plateaus are often attributed to extrinsic factors, such as reduced exposure to hazards, leaving open the question of whether mortality deceleration also occurs on the intrinsic, physiological level. Since AEGIS models intrinsic causes of death separately from extrinsic causes, such as starvation, it can disentangle intrinsic mortality from extrinsic (Fig 4e) enabling us to study deceleration of mortality at the intrinsic level as well.

Much like studying late-life mortality trajectories, AEGIS serves as a tool for investigating at what age aging starts. Molecular evidence hints that aging might start as early as birth or during early life [44], while theoretical models of resource allocation suggest different ages as starting points [45], often highlighting sexual maturity as an inflection point in life history. In our example, evolved population starts “aging”, measured as the acceleration of age-specific population mortality, at the onset of sexual maturity. To note, this result is not “imposed” as a parameter, but is rather emerging during simulations. More specifically, even though observed mortality rates seem to rise a few age classes after reaching maturity (Fig 4b), analyzing the intrinsic mortality data we found that aging starts right at the point of maturity itself (Fig 4e). A more sensitive detection of the onset of aging is possible looking at intrinsic mortality since the baseline is lower than the baseline of total mortality. We hypothesize that such an aging “schedule” is driven by strong selective pressures on variants that increase survival in the pre-reproductive stage. However, not all simulations in AEGIS will necessarily exhibit a flat pre-reproductive mortality. Increasing or decreasing pre-reproductive mortality (S1 File Supplementary Information: Juvenile mortality) might evolve under specific evolutionary scenarios.

AEGIS reveals within-population life history variation

Even though life history traits are often analyzed and represented at the population level, substantial variation in survival and reproductive outcomes may exist between subpopulations or even individuals. One approach to examining these differences is through the analysis of survivorship curves, which can exhibit considerable variation across species [46]. More specifically, the so-called rectangularization of the survivorship curve [47] can indicate the level of life history diversity, driven by either genetic differences or extrinsic factors.

In the simulations shown above, survivorship curves display a convex shape (Fig 4a, type II-III [48]), corresponding to high variance in age at death and rarity of survival into late life (Fig 2b). We hypothesized that this pattern was driven by high extrinsic mortality due to starvation (Fig 3b) generated by intense reproduction and subsequent overcrowding. Leveraging the ability of AEGIS to disentangle extrinsic factors of mortality, such as starvation, from genetically encoded intrinsic factors, we tested our hypothesis by replotting the survivorship curves considering only environment-independent intrinsic mortality. This differentiating analysis revealed a markedly rectangular survivorship curve (Fig 4d, type I [48]), implying that the intrinsic survival potential among individuals is highly homogenous and that extrinsic stochastic hazard drives lifespan variation. This finding may parallel the secular trend of rectangularization documented in human survivorship curves, which has similarly been suggested to result from reductions in extrinsic mortality (through medical advancements and decreased physical hazards) [49], rather than from changes in intrinsic mortality patterns.

Besides lifespan, life history traits can exhibit significant diversity within a population, which is crucial for the resilience, productivity, and evolvability of populations [50–52]. However, empirical studies of life history at the population level often face challenges in measuring within-population diversity or detecting the presence of exceptional subpopulations, such as centenarians [53]. AEGIS provides direct access to individual-level intrinsic life history traits, revealing considerable standing genetic variation in the simulated populations (S3b and S3c Fig), even when variant phenotypes are rare. High genetic variation suggests that rapid adaptation may be possible, though intense bottlenecking could allow poorly adapted phenotypes to dominate.

AEGIS simulates population dynamics

Previously, in Figs 3 and 4, we presented demography as static cross-sectional snapshots; however, populations in AEGIS are inherently dynamic, as are natural populations (Fig 5). These dynamics arise from the interplay between evolved intrinsic traits, i.e., mortality and fertility, and extrinsic factors, such as limited resources. Population dynamics critically influence the evolution of life history traits [9,54,55]. In AEGIS, population dynamics emerge spontaneously but can also be explicitly modified to investigate their causal relationship to life history evolution.

Periodic extrinsic pressures are recognized as influential forces in shaping the evolution of aging, although their effects can be challenging to model accurately [56]. We implemented a module mimicking abiotic factors of mortality which are periodic (Fig 5c). We can study continuous (S4a-c Fig) and intermittent pressures (Fig 4e and 4f) resulting in cyclical and periodic fluctuations of the size of the evolving population.

How populations respond to resource limitation strongly shapes demography and, consequently, the evolution of life history [57]. Population response to resource availability is marked by significant physiological diversity [58–68], which manifests in varying mortality patterns at the demographic level—patterns that we aim to capture within the AEGIS framework. By default, the mortality in response to starvation is delayed and increases gradually (Fig 5b); however, the delay can be adjusted using input parameters. In an illustrative, extreme scenario with no delay, the population shrinks immediately down to carrying capacity when confronted with starvation (S4d Fig).

AEGIS supports population genetics analyses

AEGIS offers access to simulated genotypic data across individuals and simulation time (Fig 6a). While the simulated data do not include molecular information or references to specific genes, they describe a set of variants with defined effects on life-history traits. This allows users to identify loci influencing the evolutionary dynamics of these traits.

Simulated genotypic data can be summarized and analyzed using population genetic methods, which are vital for extracting insights from large datasets. To explore the evolutionary forces—such as directional selection, balancing selection, or neutral processes—driving life-history evolution in each simulation, one might employ the site frequency spectrum (SFS). The SFS summarizes allele frequencies within a population and is widely used to infer evolutionary processes, including selection, genetic drift, and demographic changes. In Fig 6b, we present the SFS for the ancestral population at generation 1000 and the derived population at generation 10,000. Visual inspection reveals that the SFS lacks the characteristic U-shaped curve, suggesting the absence of a selective sweep within this timeframe. Additionally, the absence of an excess of intermediate-frequency alleles suggests that balancing selection did not occur. These patterns indicate that evolution may have proceeded primarily under neutral drift or weak background selection.

Given that the phenotypic effects of each variant are explicitly defined in AEGIS, this information can be incorporated into site frequency spectra analyses. For instance, it is possible to examine the age at which a variant exerts its effect (Fig 6c) or whether the variant has beneficial or deleterious consequences (Fig 6d). This allows for more detailed investigations into the evolutionary dynamics of trait-associated variants, providing deeper insight into how these variants shape life-history evolution.

AEGIS: A practical example

To illustrate the application of AEGIS for in silico experimental evolution and its potential capabilities, we replicated and extended the foundational experiment on evolution of aging conducted by Rose [41] (from here on referred to as Rose’s experiment), which tested whether imposing delayed reproduction could drive the evolution of postponed senescence in laboratory-maintained Drosophila populations.

We configured the experimental settings – parameterization, initialization, and termination – to best replicate the conditions of Rose’s study (Fig 7a). Our simulation involves the evolution of two populations with non-overlapping generations – following Rose’s original notation, a base population (B) and a population (O) in which only the oldest individuals reproduce. Both populations share identical simulation parameters, with the only difference being that in the O population, individuals can reproduce beyond day 14 (up to 50 days) and the eggs laid later were preferentially carried over into the next generation. In Rose’s study, population B was maintained for 50 generations and population O for 15 generations, whereas both of simulated populations were run for approximately 50 generations (equivalent to 1000 steps, with the number of generations contingent on cohort lifespan). We initialized the B and the O population using a pre-evolved population that evolved under B conditions but for a significantly longer time (over 2000 generations) to prevent initialization artifacts (see Methods: Experimental Design).

The simulations show that population O evolves both an increased median and maximum lifespan (Fig 7b), suggesting that evolutionary pressures favor the evolution of extended lifespan, consistent with the experimental findings of Rose.

Although our experiment indicates that selection for late-life reproduction contributes to the evolution of delayed senescence, it remains unclear whether other conditions of the experimental evolution are necessary as well. Importantly, several aspects of the experimental setup were not varied, meaning that the experiment does not test whether these aspects are essential for the evolution of delayed senescence.

Using AEGIS, we can examine the importance of different conditions for the evolution of delayed senescence. One critical difference between laboratory and wild-like scenarios is the high level of extrinsic mortality present in the wild. It has long been posited that high extrinsic mortality obscures the selection for late-life traits in natural populations. We can assess how our findings generalize to such environments by simulating a wildlife scenario through two key modifications: transitioning from non-overlapping to overlapping generations and introducing limited resources, which impose extrinsic mortality on the population. In the wild-like context, we do not study the evolution of senescence per se; instead, we observe how evolved populations perform under these wild-like conditions. Our results show that the evolved populations in wild-like conditions exhibit minimal differences in survivorship between the O and B populations, with most of the survivorship curves being nearly identical (Fig 7c) and only slight variation in late-life survival (Fig 7d). This suggests that the fitness advantage of longer lifespan observed in the laboratory is not strongly expressed in wild-like environments and indicates that such adaptations are unlikely to evolve under wild conditions. This highlights how AEGIS can effectively generalize findings which may otherwise be difficult to achieve.