Skip to main content
Open Access

The Demographic Consequences of Adaptation: Evidence from Experimental Evolution

Abstract

The process of adaptation toward novel environments is directly connected to the acquisition of higher fitness relative to others. Such increased fitness is obtained by changes in life history traits that may directly impact population dynamics. From a functional perspective, increased fitness can be achieved through higher resource use or more efficient resource use, each potentially having its own impact on population dynamics. In the first case, adaptation is expected to directly translate into higher population growth. In the second case, adaptation requires less energy and hence may lead to higher carrying capacity. Adaptation may thus lead to changes in ecological dynamics and vice versa. Here, by using a combination of evolutionary experiments with spider mites and a population dynamic model, we investigate how an increase in fecundity (a validated proxy for adaptation) affects a population’s ecological dynamics. Our results show that adaptation can positively affect population growth rate and either positively or negatively affect carrying capacity, depending on the ecological condition leading to variation in adaptation. These findings show the importance of evolution for population dynamics in changing environments, which may ultimately affect the stability and resilience of populations.

De gevolgen van adaptatie op de demografie: bewijs verkregen uit experimentele evolutie

Het proces van adaptatie aan nieuwe omgevingen is rechtstreeks gekoppeld aan het verwerven van een hogere fitness. Veranderingen in fitness worden bereikt door wijzigingen in de levensgeschiedeniskenmerken zoals overleving en reproductie. Deze kenmerken sturen de populatiedynamieken op een directe manier. Vanuit een functioneel oogpunt kan een fitnesstoename bereikt worden door een verhoogde voedselinname of een efficiënter gebruik ervan. Wanneer de inname van voedsel verhoogt wordt er verwacht dat deze inname zal leiden tot een hogere populatiegroei. Wanneer efficiënter met voedsel wordt omgesprongen zal adaptatie minder energie vergen en kan dit bijgevolg leiden tot een hogere draagkracht. Adaptatie kan dus veranderingen in de ecologische dynamieken veroorzaken en omgekeerd. We rapporteren inzichten van een experimentele evolutie met spintmijten als modelorganisme. Via een populatiedynamisch model testen we of een toename in fecunditeit (een betrouwbare proxy voor adaptatie) de ecologische dynamieken verandert. Onze resultaten tonen aan dat adaptatie een positieve invloed heeft op populatiegroei, en dat het zowel een positief als een negatief effect kan hebben op de draagkracht afhankelijk van de ecologische omstandigheden die tot de variatie in adaptatie geleid hebben. Deze bevindingen tonen het belang van evolutie voor populatiedynamieken in veranderende omgevingen, met mogelijke gevolgen voor de stabiliteit en weerbaarheid van populaties.

Online enhancements:   supplemental PDF.

Introduction

Adaptation to novel environments involves changes in life history traits, such as number of offspring, adult mortality, and developmental rate (Stearns 1976). These evolutionary changes may then induce a feedback on demographic properties of populations (ecological dynamics; Hendry 2016, 2019; Govaert et al. 2019), such as population growth and the equilibrium population size or carrying capacity. High population growth induces, for instance, faster recovery after a population collapse (Turcotte et al. 2011, 2013). However, fast growth may also cause chaotic dynamics and hence higher instability due to overcompensation (Best et al. 2007). In addition, higher carrying capacities sustain larger populations, usually with higher evolutionary potential (more de novo mutations under stable mutation rates and genetic recombination) and lower chances for adaptive decline (lower chance for inbreeding depression and genetic drift; Olson-Manning et al. 2012; Yates and Fraser 2014; Abrams 2019). Investigating these eco-evolutionary feedbacks is therefore crucial, as ecological dynamics may influence the stability and resilience of populations (Strauss 2014; Hendry 2016, 2019).

If evolutionary changes feed back on the population growth rate (i.e., better-adapted populations have a higher population growth rate) and resources are restricted, then intraspecific competition will benefit individuals with a higher resource efficiency and hence a lower need of resources. Higher resource efficiency can be obtained through lowered handling times or increased metabolic efficiency, such as the evolution of detoxification mechanisms (Després et al. 2007; Van Leeuwen and Dermauw 2016; Dermauw et al. 2018; Rane et al. 2019). Another advantageous strategy against resource limitation might be to quickly acquire resources before resource limitation arises, which would favor higher resource use instead of more efficient resource use. The spatial context (e.g., resource availability and immigration rates or the number of individuals moving from one patch to another in a landscape) may further tune the different resource strategies. High immigration rates or high connectivity could, for instance, increase competition and hence result in resource limitation in patches with a positive immigration-emigration balance.

Evolutionary changes may also feed back on carrying capacity. Depending on the evolved resource strategy, one may expect adaptation to result in lower or higher equilibrium population sizes, which we refer to as carrying capacities. A lower carrying capacity is expected when individuals increase their per capita resource use and do not entirely invest the surplus energy in offspring but instead invest it in, for instance, competition. More efficient resource use can result in an increase in carrying capacity: the higher resource efficiency decreases the per capita need for resources, allowing more individuals to subsist on the same amount of resources.

Although the idea of adaptation influencing carrying capacity already has a long history (Haldane 1932; Roughgarden 1971; Heckel and Roughgarden 1980), empirical evidence is scarce (except for predator-prey systems; e.g., Yoshida et al. 2003; Hairston et al. 2005; Becks et al. 2010, 2012; Hiltunen and Becks 2014; Hiltunen et al. 2014). A recent theoretical study of multiple consumer-resource models suggests that populations should usually decrease in population size following adaptation (Abrams 2019). In that study, the main cause of the decreasing population size is an increase in resource uptake by the consumer, which leads to a decrease in the growth rate of the resource due to a lower availability of resources or a trade-off between growth and defense of the resource. Another possible explanation given by Abrams (2019) is the trade-off between resource uptake and resource conversion efficiency.

In this study, we fitted available data from an evolutionary experiment (Alzate et al. 2017, 2019) to a population model to examine how differences in adaptation to a novel environment affect the population growth factor (geometric rate of increase; Pachepsky et al. 2008) and/or carrying capacity. The use of experimental populations allows us to study eco-evolutionary dynamics at a longer timescale, where populations approach equilibrium in a more controlled environmental context than observational studies (Schoener 2011; De Meester et al. 2019). In these experiments, populations of the two-spotted spider mite (Tetranychus urticae Koch, 1836) adapted to a novel challenging host plant (tomato), with migration rates, resource availability, and interspecific competition being manipulated (fig. 1). The results from the competition treatment are not used here because the interspecific competitor may have an effect on population dynamics, and its population size was not recorded. The manipulation of both immigration rates and resource availability allowed us to validate predictions about the rate or extent of (mal)adaptation toward the novel environmental conditions that are known to impose strong selective forces on the species’ life history (Alzate et al. 2017, 2019). Because of the different levels of acquired adaptation in the experimental metapopulations, the link between adaptation level with the population dynamics could be explicitly tested. We fitted a single-species Ricker-type model to the data from these experiments. We chose a single-species model instead of a plant-herbivore model because the resources were refreshed on a biweekly basis and hence considered to be constant (but see “Methods”). We define adaptation as the state of organismal performance in a specific environment (Brady et al. 2019) after several generations of evolution via natural selection in that environment. The two-spotted spider mite is an r-selected species, and hence fitness maximization leads to higher reproduction in the absence of competition (MacArthur and Wilson 1967), making fecundity a good proxy for fitness and hence adaptation in our model organism. Other fitness-related traits, such as survival and developmental rate, have a lower evolutionary potential, are more stochastic, and are harder to precisely measure in our model organism (Magalhães et al. 2007). We therefore used the evolved fecundity at the individual level (measured under common garden conditions and without competition) after 20 generations of evolution in different experimental environments as a signal of adaptation to each experimental environment (see “Methods”) to explore the consequences of individual performance for population dynamics. We anticipate that increases in fecundity during adaptation will increase the population growth factors. Our expectations for the influence of adaptation on carrying capacity are less straightforward, as they will depend on per capita resource use and the amount of energy invested in offspring.

Figure 1. 
Figure 1. 

Overview of experiments. On the x-axis the different levels for immigration rate α are illustrated, while the y-axis shows resource availability c. Resource availability is not the same as the number of physical plants, which is shown by the number of tomato plants in the figure. Plants in the immigration load experiment (gray) are older/larger and can sustain larger population sizes, so one plant in the immigration load experiment is similar to 2.43 plants in the island biogeography experiment (white; for correlation, see fig. S3). In the island biogeography experiment, three levels of resource availability (one, two, and four tomato plants per island) and three different immigration rates (one mite every 2 weeks, one mite per week, and two mites per week) were used. The immigration load experiment had one level for resource availability and four different immigration rates (two, three, five, and 10 mites per week). The adaptation levels per treatment are visualized in figure S1.

Methods

Study Species

The two-spotted spider mite Tetranychus urticae Koch, 1836 (Acari, Tetranychidae), is a cosmopolitan generalist herbivore (Gotoh et al. 1993; Bolland et al. 1998) that is regarded as a suitable model species for evolutionary experiments because of its high fecundity (1–12 eggs/day), short generation time (11–28 days), and small body size (∼0.4 mm in length; Gould 1979; Fry 1989; Agrawal 2000; Magalhães et al. 2007; Bonte et al. 2010; Alzate et al. 2017, 2019; Bisschop et al. 2019, 2020). Response to selection in the two-spotted spider mite has been previously observed after just five generations (Agrawal 2000), and adaptation was reported after 15 to 20 generations of selection to a novel host (Magalhães et al. 2009; Alzate et al. 2017; Bisschop et al. 2019). We used the London strain of T. urticae for experimental evolution. This strain was initially collected from the Vineland region in Ontario, Canada (Grbić et al. 2011), and is known for its high-standing genetic variation (Wybouw et al. 2015). The laboratory strain has been maintained on the bean plant Phaseolus vulgaris ‘Prelude’ for more than 5 years.

Experiments

Two previously performed evolutionary experiments were used in this study (Alzate et al. 2017, 2019). Both experiments investigated the spatial context of local adaptation from an ancestral population on the bean plant P. vulgaris ‘Prelude’ to the tomato plant Solanum lycopersicum ‘moneymaker.’ Although the idea of exploring the demographic dynamics following adaptation was planned beforehand, the specific hypotheses for the population dynamics model were fine-tuned after performing the experiments. Tomato plants are challenging for spider mites because of their induced responses through phytohormones (e.g., ethylene, salicylic acid, and jasmonic acid) and glandular trichomes (Lucini et al. 2015; Godinho et al. 2016).

In the first experiment (Alzate et al. 2019), the joint effect of immigration rate (0.5, one, and two mites/week) and resource availability (one, two, and four plants per island) was tested on the process of adaptation to tomato plants (fig. 1). Hereafter, this experiment will be referred to as the “island biogeography experiment.”

The second experiment (Alzate et al. 2017) examined the interplay between interspecific competition with a congeneric species and immigration rates (two, three, five, and 10 mites/week) with respect to adaptation to tomato plants. In our study, we considered only the effect of different immigration rates of the treatment without interspecific competition and not the interspecific competition treatment itself in order to focus on the population dynamics of T. urticae in isolation. We refer to the second experiment as the “immigration load experiment,” reflecting the fact that the immigration rates were higher than in the island biogeography experiment (fig. 1). In comparison to the island biogeography experiment, here only one tomato plant was used per island, and hence there were no differences between islands in terms of resource availability.

Both experiments started with adult female mites coming from the same ancestral population reared on bean plants and therefore not adapted to tomato plants. The two experiments were simultaneously performed within the same climate room, providing identical climate conditions (25°C±0.5°C and a 16L∶8D photoperiod). The experiments ran for about 20 generations (~13 days per generation). In total, there were five replicates for each of the nine treatments (three levels for immigration rate and three levels for resource availability) in the island biogeography experiment and seven replicates for the four treatments (four levels for immigration rate) in the immigration load experiment. Plants were refreshed every 2 weeks by placing all leaves with mites from the old plants to new, fresh tomato plants. We note that the “old” plants were 2 weeks older and larger than the fresh plants and consequently had larger carrying capacities. While the plant refreshment procedure was meant to provide constant resources, they were evidently not entirely constant, and hence plant refreshments may have resulted in population fluctuations. However, we minimized these population fluctuations caused by the biweekly plant refreshment procedure by using only the population sizes assessed just before the plant refreshment in those weeks where plants were refreshed and on similar weekdays in the other weeks for our analyses. This resulted in a weekly count of the population sizes (23 and 22 population counts for the island biogeography and immigration load experiments, respectively). In the island biogeography experiment, population counts are missing for 2 weeks. We accounted for this in the model as explained below.

Alzate et al. (2017, 2019) measured the fecundity after 20 generations of putative adaptation to the novel host plant. To reduce juvenile and maternal effects (Magalhães et al. 2011; Kawecki et al. 2012) on fecundity, they transferred the inseminated female mites separately onto bean leaf discs for two generations (common garden). After these two generations, the fecundity on tomato was assessed by transferring a single female offspring from this last common garden generation to a tomato leaf disc. For each female, the total number of eggs after 6 days was counted. We use fecundity in our calculations as a proxy for adaptation level because this trait has high evolutionary potential in T. urticae compared with survival or developmental rate (Magalhães et al. 2007). The adaptation level is obtained by dividing the number of eggs per female by the maximum number of eggs found on the tomato from a single female in any of the treatments (with 1 being the highest obtained adaptation level). Per experimental population, individual levels of adaptation were averaged and used in the population dynamic model (for the attained adaptation levels in the two experiments, see fig. S1).

Population Dynamics Model and Statistical Analysis

The goal of this study is to examine whether the level of adaptation to a new habitat results in differences in carrying capacity K (i.e., the equilibrium population size where Nt+1 is equal to Nt) and/or population growth, which we here measure by the growth factor or geometric rate of increase λ (λ=er, with r being the population growth rate; Pachepsky et al. 2008).

Both λ and K were estimated from the weekly measured population sizes (Nt) using a discrete-time model of population dynamics based on the Ricker model (Ricker 1954; eq. [1]) that incorporates negative density dependence (see below for the resulting recurrence equation for population size). We chose a single-species model instead of a plant-herbivore model because the resources were refreshed every 2 weeks. The Ricker model was preferred over Beverton-Holt as a basis because most spider mite populations had an initial steep growth followed by oscillations after an overshoot. Given that the Ricker model is known for overcompensation in population dynamics, it seems to be well suited for our data set. The Ricker model is mathematically represented as

(1)Nt+1=Nter(1Nt/K)=λNteβNt,
where λ=er is the growth factor (Pachepsky et al. 2008) and β=r/K.

As the Ricker model does not include immigration, we modified equation (1) to include an immigration rate α similar to the one in our experiments (every week a fixed number of immigrants was added to the population; eq. [2]). Because the mites were transferred directly after the weekly count, the immigrants can be added to the measured population size, and they are able to reproduce along with the resident population.

We counted only the adult females for the ecological dynamics in the experiments because they provide a good proxy of the entire population size due to overlapping generations. Females are long living and hence give a more integrated and balanced snapshot of population size. Ecologically, they consume most of the resources. However, males and juveniles may also affect the density-dependent term β in the equation. We therefore multiplied Nt in the density-dependent term by a factor of 1.3 to account for the exclusion of juveniles and adult males in the total population size. We chose a factor of 1.3 because adult females contribute to about 26% of the entire population and consume about nine times more resources than males and juveniles, which implies that adult females eat 76% (=1/1.3) of the total daily consumed resources (De Roissart et al. 2015; Bonte and Bafort 2019). Sensitivity analysis showed that our results were robust to changes in the factor of 1.3 in the density-dependent term (fig. S2). Incorporating all of the above-mentioned considerations, our recurrence equation for population size becomes

(2)Nt+1=λ(Nt+α)eβ(1.3Nt+α).
In the original Ricker model (eq. [1]), K is defined as the population density for which Nt+1=Nt. We constructed the density-dependent term β in the same way for the modified Ricker model (eq. [2]) as was done in original Ricker model (we set Nt+1 and Nt equal to K′ in eq. [2] and log transformed both sides of the equation to isolate β):
(3)β=ln(λ)+ln(K+α)ln(K)1.3K+α.
We used K′ instead of K because the number of plants and hence the carrying capacity differed in the different treatments of the experiment (some islands consisted of one, two, or four plants in the island biogeography experiment). We therefore defined K as the carrying capacity of adult females per plant, and the total K′ is then the product of the number of plants, n, and K (K=nK).

We assumed a linear dependence of λ (with intercept a0 and slope a1) and K (with intercept b0 and slope b1) on the level of adaptation f (continuous variable). Setting the slope to zero makes λ and K independent of f.

We quantified the level of adaptation f at the end of the evolutionary experiment (eqq. [4], [5]), which occurs after 20 generations of putative adaptation to tomato plants. Therefore, the influence of the reached level of adaptation on demography will be weaker at the beginning (before adaptation occurred) than at the end of the experiment. To account for such a time-dependent effect of adaptation on λ and K, we implemented two extra parameters within a sigmoid function. These two parameters are s1 (the shape of the sigmoid function) and s2 (the reflection point of the sigmoid function), as they define the influence of time on adaptation f; they are the same for both λ and K.

Because of the differences in plant age between the experiments, the population carrying capacity Ki differs between the experiments: plants in the immigration load experiment were 1 week older and probably provided more resources than the plants in the island biogeography experiment. To be able to use the same equation for both experiments and to accommodate this difference, we introduced a multiplier parameter c. The value of c was estimated from the raw data from the island biogeography experiment, where we regressed the mean population size at the plateau phase against the resource availability (i.e., the number of plants). The resource availability of the immigration load experiment is estimated from this obtained regression line (fig. S3). The value of c was 1 in the island biogeography experiment and 2.43 in the immigration load experiment. All of these considerations lead to the following dependencies of λ and K on f (and time t):

(4)λ=a0+a111+es1(ts2)f,(5)Kil=cKib=c(b0+b111+es1(ts2)f),
where subscript “il” denotes the immigration load experiment and subscript “ib” denotes the island biogeography experiment. We performed a sensitivity analysis for the value of c that showed no strong influence on the preferred model or on the population growth factors in the immigration load experiment (fig. S4). There was an influence of c on carrying capacity, but this is to be expected given that b0 and b1 need to be smaller when c increases, following equation (5) (fig. S4). To allow for stochasticity in the count data, we used a negative binomial error distribution on Nt+1 with a mean described by the above-mentioned equation for Nt+1 (eq. [2]) and a dispersion parameter, σ (i.e., the parameter “size” in function dnbinom from the R package stats). For infinite σ, the negative binomial reduces to a Poisson distribution. The likelihood L of the model is then a product of the negative binomials across all time points from the start at zero until the final time point T and across all replicates:
(6)L=replicatest=0T1NB(Nt+1,σ;Nt,a0,a1,b0,b1,c,s1,s2).
For the two missing counts in the island biogeography experiment, this likelihood is summed over all possible population sizes in the missing weeks.

To test the impact of the adaptation level on the population growth factor and/or the carrying capacity, we estimated a0, a1, b0, b1, c, s1, s2, and σ with maximum likelihood, using a list of 100 initial parameter sets to find a global likelihood optimum (initial parameter sets; table S1). We also provide an estimate of the error in the parameter estimates. Uncertainty in parameter estimates is usually estimated by measuring the standard deviation of a Gaussian approximation of the log-likelihood surface at the optimum. This approximation, however, is not possible if the estimations are at the borders of the parameter domain and not at a local optimum (where the derivatives of the likelihood function with respect to the parameters are zero). We therefore chose to provide an interval of parameter estimates for which the log likelihood is two units lower than the maximum likelihood (indicated by brackets in “Results”). Two units was selected because it is commonly used as a significant difference between models that differ in 1 df (one parameter).

We compared six models (M1–M6; fig. 2; table 1) differing in which parameters were held fixed (explained below) and selected the best model on the basis of the highest Akaike information criterion (AIC) weight. M1 is the null scenario, where evolution does not change the per capita resource use or the fecundity, and hence no change in λ (a1=0) or K (b1=0) is expected. M2 and M3 are scenarios where more efficient use of resources leads to an increase in fecundity but does not affect the per capita resource use. Therefore, we only expect a change in λ and not in K (b1=0). The main difference between M2 and M3 is that the population growth factor λ depends entirely on the level of adaptation in M3 (a0=0, so λ=a1sf). There are two different scenarios for each of the models M4, M5, and M6 depending on the sign of the relationship between carrying capacity and adaptation (see fig. 2). The main differences between the three models are that in M4 the increase in λ is independent of the level of adaptation (a1=0), while it depends on adaptation in both M5 (a0=0) and M6 (none of the parameters are fixed). The distinction between M5 and M6 is that λ depends entirely on the adaptation level in M5 but not in M6 (in analogy to M2 and M3). We note that per capita resource use was not explicitly tested in our study; we include it only as a theoretical underlying mechanism. We fitted the same model to the data for both experiments separately as well as to the entire data set. In the latter data set, we optimized the parameters while keeping some parameters equal between both experiments (see below) to allow differences between the island biogeography experiment and the immigration load experiment, leading to a total of 106 different models (table S2). We studied models with all seven parameters in common between experiments (table 1), no parameters in common, or only one parameter in common. Furthermore, we explored combinations of common parameters: the two parameters for K, the two parameters for λ, all four parameters for K and λ, parameters for K and λ with one of the sigmoid parameters equal in both experiments (s1 or s2) or both sigmoid parameters in common, parameters for K and λ with the dispersion parameter σ also in common, and parameters for K and λ with both the dispersion parameter σ and s1 in common. Last, we also tested models with none of the parameters for K and λ in common but with the parameters for the sigmoid (s1 and s2) and the dispersion parameter in common. These different models and their results are provided in table S2.

Figure 2. 
Figure 2. 

Tested models with their respective hypotheses after evolution to novel conditions and how the evolution changes population growth factor λ and carrying capacity K. The missing part of the leaf represents per capita resource use, while the circles indicate the number of eggs (fecundity) per individual. The top row shows the initial situation before evolution to novel conditions, while the middle and bottom rows indicate the possible scenarios resulting from evolution to novel conditions. The combination of resource use and fecundity results in either positive or negative effects on λ and K (shown as x- and y-axes in the little graphs). The different models are shown as M1–M6. M1 is the neutral model, where resource use and reproductive output are unaffected by the novel environment. M2 and M3 are the resource efficiency for fecundity, partially, hypothesis, where more efficient resource use does not affect K but leads to an increase in fecundity and hence a positive effect on λ (M2 with intercept and M3 without intercept). M4–M6 are divided between two different hypotheses depending on per capita resource use; lower resource use due to higher efficiency (the resource efficiency for fecundity hypothesis) resulting in higher K or higher resource use (resource use for fecundity hypothesis) resulting in lower K. The differences between M4, M5, and M6 are the influence of evolution on λ: no effect (M4), a positive effect without an intercept (M5), or a positive effect with an intercept (M6). We did not assess per capita resource use; we consider only the theoretical underlying mechanism.

Table 1. 

The six models analyzed with their estimated parameters for growth factor λ and carrying capacity K

ModelEstimated parametersExplanationScenario (in fig. 1)
M1a0, b0, s1, s2, σAdaptation does not influence λ and K (a1 = b1 = 0)Neutral model
M2a0, a1, b0, s1, s2, σAdaptation influences λ but not K (b1 = 0)Resource efficiency for fecundity, partially
M3a1, b0, s1, s2, σAdaptation influences λ entirely (a0 = 0; no intercept) but not K (b1 = 0)Resource efficiency for fecundity, partially
M4a0, b0, b1, s1, s2, σAdaptation influences K but not λ (a1 = 0)Resource efficiency for fecundity/resource use for fecundity
M5a1, b0, b1, s1, s2, σAdaptation influences λ entirely (a0 = 0; no intercept) and KResource efficiency for fecundity/resource use for fecundity
M6a0, a1, b0, b1, s1, s2, σAdaptation influences λ and KResource efficiency for fecundity/resource use for fecundity

Note.  Shown are parameters for population growth factor λ (eq. [4]) and carrying capacity K (eq. [5]). σ is an indication of the amount of overdispersion. For infinite σ, the negative binomial reduces to a Poisson distribution. The sigmoid function accounts for a changing effect of adaptation on λ and K in time with s1 (the shape of the sigmoid function) and s2 (the reflection point of the sigmoid function). The last column links the models to the scenarios in figure 1.

View Table Image

Additional Analyses

We performed additional analyses to further investigate the puzzling results of the influence of adaptation level on carrying capacity K (i.e., b1). We explored the possibility of obtaining both a negative and a positive influence of adaptation on carrying capacity via the different ecological conditions during adaptation. We subdivided the data set by the ecological conditions during adaptation (the different immigration rates and the resource availability) and ran M6 (with no fixed parameters) on these different data sets. The obtained value for b1 was then plotted against the fecundity from the different treatments. Error values for b1 were estimated as explained above, and for the fecundity data we used the same quantiles as for the boxplots in figure S1 (0.32 and 0.68). For the sake of completeness, we performed the same additional analysis for the population growth factor λ. To ascertain that our results on the subdivided data sets were not affected by the linear correlation of resource parameter c between both experiments (fig. S3), we performed a sensitivity analysis investigating M6 under different values of c (2.43, 3.5, 4, 4.5, 5, 7.5, and 10), starting from the initial 100 parameter sets as well as the obtained values from the previous resource constant. This is necessary given the contrasting results for the influence of adaptation level on carrying capacity between both experiments: if the carrying capacity of one plant is lower in one experiment, we cannot assume a linear correlation. The results of the sensitivity analysis showed that the higher the value of the resource constant c gets the closer the value for b1 gets to zero, which means that the slope of the correlation between adaptation and carrying capacity becomes less steep (also visible in fig. S4). Overall, we obtained the same trend as was found in figure 3 (positive or negative correlation between adaptation level and carrying capacity) for the lowest two and highest immigration rates (α=2, 3, and 10), which shows the robustness of these results. Only the second-highest immigration rate (α=5) revealed contrasting results (fig. S5). Given the (almost) identical log likelihoods for the different resource constants per immigration rate, we cannot know for the immigration rate of five mites per week whether the correlation between adaptation and carrying capacity is positive or negative. We need to stress that the number of replicates was on the low side (see below), so outcomes need to be treated with caution. The numbers of replicates were seven, seven, and six for the different immigration rates in ascending order for the island biogeography experiment and four, five, two, and two replicates for the immigration load experiment in ascending order. The numbers of replicates depending on the resource availability c were two, seven, and 11 for the island biogeography experiment and 13 for the immigration load experiment.

Figure 3. 
Figure 3. 

Relationship between level of adaptation and carrying capacity after subdivision of the data set by ecological condition: immigration rate α (A) and the number of plants n (B). On the x-axis the level of adaptation is presented, which is based on the number of eggs after 6 days during the fecundity assessments. The error bars are the quantiles for the boxes as visualized in figure S1 (0.32 and 0.68). On the y-axis the estimation of parameter b1 is given (which is the coefficient in the relationship between carrying capacity and fecundity) with the calculated errors. The numbers are immigration rates α and number of plants n, while the different colors represent the two experiments (blue for the immigration load experiment and red for the island biogeography experiment). The results are provided in table S3.

All analyses were performed in R (ver. 4.0.2) using the optimizer function from DDD (ver. 4.4.1; Etienne and Haegeman 2019), subplex (ver. 1.6; King and Rowan 2020), ggplot2 (ver. 2.3.3.3; Wickham 2016), and ggpubr (ver. 0.4.0; Kassambara 2018).

Results

General Results

The population dynamics were generally characterized by an initial growth phase, followed by an overshoot and an oscillatory plateau phase (fig. 4). The demography thus resembles Ricker dynamics.

Figure 4. 
Figure 4. 

Ecological dynamics per population with the model estimation of M6 for the island biogeography experiment (A) and the immigration load experiment (B; see table S3 for parameters of both models). The number of adult females is plotted against the time in days (points are the empirical data, and curves present the Ricker-type model fitted to the data). Each color per plot represents an individual population (immigration rate α is shown per column, and the rows indicate the number of plants). Fecundity assessments on tomato plants for small islands with 0.5 and two mites per week failed.

When considering the data sets of the experiments separately, the model with adaptation affecting both population growth factor λ and carrying capacity K (M6; table 1) performed best for the two data sets (AIC weight=0.67 for the island biogeography experiment; AIC weight=0.57 for the immigration load experiment; table S3). The models performed relatively well, as the difference between estimated and measured population size (Nt+1) varied around zero (fig. S6).

We found a significant positive correlation between the level of reached adaptation f and the population growth factor λ for the island biogeography experiment (a0=1.49 [1.37; 1.61] and a1=0.37 [0.18; 0.58]) and the immigration load experiment (a0=0.98 [0.84; 1.13] and a1=1.11 [0.85; 1.41]; fig. 5A, 5D). The second-best model for both experiments based on the AIC weights (0.26 and 0.43, respectively; table S3) assumed adaptation to impact growth factor only (M2; table 1). The positive correlation between adaptation and population growth factor was thus overall strongly supported (combined AIC weight=0.93 for the island biogeography experiment; combined AIC weight=1.00 for the immigration load experiment; table S3).

Figure 5. 
Figure 5. 

Model output from the best model (M6) for population growth factor λ, carrying capacity K, and the sigmoid curve in the island biogeography experiment (AC) and immigration load experiment (DF). The x-axis presents the level of adaptation (A, B, D, E), which is based on the number of eggs after 6 days during the fecundity assessments, or the time in days (C, F). The y-axis represents population growth factor λ (from parameters a0 and a1 in A and D), carrying capacity K (from parameters b0 and b1 in B and E), and the magnitude influence of adaptation (from parameters s1 and s2 in C and F). The solid line shows the parameter estimates. The gray zone indicates the error on the intercept and slope, which are values for which the log likelihood is two units less than the maximum likelihood.

The relationship between reached adaptation level f and carrying capacity K differed between both experiments. In the island biogeography experiment a positive relationship was found with the level of adaptation (b0=17.46 [15.93; 19.28] and b1=6.19 [4.41; 8.34]), while this relationship was negative for the immigration load experiment (b0=26.12 [24.20; 28.59] and b1=4.29 [−5.71; −2.34]; fig. 5B, 5E). In this light, it is important to recall that the second-best model for both experiments, M2, assumes no correlation with carrying capacity K (table S3).

The combined data set revealed the same results (tables S2, S4): a positive correlation between fecundity and population growth factor λ and a condition-specific correlation with carrying capacity K (fig. 5). The three most supported models had none of the parameters equal between the two experiments or only the parameter for the shape of the sigmoid function, s1.

Because adaptation tunes in only after some generations of experimental evolution, we assume no influence of adaptation from the start on population dynamics. The influence of adaptation on the ongoing population dynamics may be gradual or abrupt (small vs. large values for s1, respectively) and may vary in timing (the reflection point s2). The different sigmoid curves obtained from the optimized parameters are presented in figure 5C and 5F. The main difference between both experiments was found in the reflection point, which occurred after 75 days ([70.10; 84.03]; table S3) in the island biogeography experiment and already after 25 days ([21.02; 27.00]; table S3) in the immigration load experiment.

Additional Analyses

After finding the above-mentioned results showing differences between the immigration load experiment and the island biogeography experiment, we investigated the influence of the ecological conditions during the experiment—namely, resource availability and the different immigration rates—on carrying capacity.

We discovered a possible effect of the different immigration rates (fig. 3; table S5): for immigration rates with on average low levels of adaptation (α=0.5, 5, or 10 mites per week) a negative relationship between adaptation and carrying capacity was found in both the island biogeography experiment and the immigration load experiment, while positive relationships were found for higher levels of adaptation (α=1, 2, or 3 mites per week). We found the highest levels of adaptation for intermediate immigration rates and lower levels for low and high immigration rates (Alzate et al. 2017, 2019), and we observed that the higher the adaptation level, the higher the influence on carrying capacity. The additional analysis on the population growth factor did not reveal any further insights based on immigration rates or resource availability; the relationship between level of adaptation and population growth factor was always positive (fig. S7).

Discussion

We investigated the influence of adaptation on the population growth and carrying capacity of experimental spider mite populations. Adaptation to a novel environment, tomato plants, increased population growth λ (fig. 5A, 5D). Populations of better-adapted individuals at the end of the experiment (after 20 generations) thus grew faster during the course of the experiment. Furthermore, we found either a positive or a negative relationship between adaptation and carrying capacity in our populations (fig. 5B, 5E). The sign of the relationship between adaptation and carrying capacity depended on the experimental conditions leading to variation in adaptation: the level of immigration and resource availability.

The population growth factor is key to population dynamics and is determined by life history traits as fecundity and survival (Tanner 1975). In this study, we investigated whether adaptation at the individual level is related to the population growth factor. It may seem trivial that higher fecundity, which is our proxy for the eventual reached level of adaptation, increases the population growth factor because one would expect this to be a logical necessity. However, fecundity was measured for individual females on single-leaf cuts in the absence of other conspecifics, while the ecological dynamics were derived from population counts on the tomato islands. It is possible that an individual has high fecundity when placed alone but not when it is placed under competition. Thus, higher fitness may result in higher resource use, which is allocated to reproduction when alone but to increasing competitive strength when under (intraspecific) competition. In Tetranychus urticae, fecundity has been shown to be a reliable proxy for adaptation in comparison to, for instance, baseline survival and development rate, which are experimentally difficult to assess with great precision (Magalhães et al. 2007). The positive relationship between adaptation and population growth factor is in line with the findings of Turcotte et al. (2011, 2013), who performed an experiment with the green peach aphid, Myzus persicae, and demonstrated that population growth during the exponential growth phase following adaptation to novel host plants was enhanced by evolution. Yet our study investigated the population dynamics not only during the exponential growth phase but also throughout the establishment and equilibrium conditions, where populations were at carrying capacity.

A larger population growth factor may indicate that populations are more stable and resilient to environmental perturbations, as they will reach the carrying capacity faster. However, high population growth may also lead to overcompensatory mechanisms under scramble competition and even population collapse (Best et al. 2007). In two-spotted spider mites we often see that colonization of novel plant resources occurs with small numbers, followed by fast growth with overexploitation and a potential overshoot (also visible in fig. 4). It is hence not straightforward to link evolved population growth factors to stability, as delayed density dependence may eventually lead to population decline.

We found puzzling results for the influence of adaptation on carrying capacity. In general, we observed a decrease in carrying capacity with adaptive evolutionary change for the immigration load experiment, while an increase was seen for the island biogeography experiment (fig. 5B, 5E). The outcome that adaptation may lead to both lower and higher carrying capacities is fascinating, especially in light of population stability and resilience. Overall, larger populations have a higher evolutionary potential (Olson-Manning et al. 2012; Yates and Fraser 2014; Abrams 2019), which may help populations to persist in changing environments. Therefore, evolution to lower carrying capacities may be disadvantageous.

This result potentially points to different hypothetical resource strategies between both experiments: some populations increased their resource use, leading to lower population sizes, while other populations evolved a more efficient use of resources, increasing their population sizes. The main ecological differences between both experiments were the different resource availabilities and immigration rates. We therefore investigated in the additional analyses whether a certain ecological condition could lead to a different resource strategy (fig. 2).

Given that the island biogeography experiment consisted overall of larger islands than the increased immigration experiment, those larger islands may have determined the final result for this experiment: an increase in carrying capacity with adaptation. We performed an additional test dividing the data set by level of resource availability to evaluate whether islands with similar resource availability between both experiments showed equal trends. We indeed found that for the populations with similar resource availability (2c2.43), higher adaptation levels meant lower carrying capacities. This suggests that more resources are used for increasing competitive strength (figs. 1, 3; table S5). In contrast, populations on small and large islands seemed to evolve resource use efficiency (figs. 1, 3; table S5). We expect a lower degree of intraspecific competition on large islands than small islands, especially with the refreshment procedure used in our experimental setup. The populations were refreshed biweekly by placing all leaves from the old tomato plants on the fresh plants. Fresh plants were always the same age, so the fresh plants were 2 weeks younger and hence smaller than the old tomato plants. This means that a population is suddenly transferred to a smaller area, which strongly increases the population density of the spider mites, potentially leading to a necessity of fast acquisition of resources, as a few individuals may take it all. Alternatively, the lower level of competition on the large islands may have provided the necessary time to evolve a strategy of resource efficiency. However, in situations where the resources are extremely limited and competition is intense, individuals may be forced to evolve a higher efficiency for acquisition and conversion of resources. To recapitulate, we hypothesize that under high intraspecific competition population evolution leads to an increase in their resource use, but under high levels of resource depletion it leads to a more efficient resource use (fig. S8).

Another explanation for the puzzling results between both experiments could be related to the different levels of immigration. Different immigration rates were implemented in both studies, ranging from one mite every 2 weeks up to 10 mites per week. The higher immigration rates in the immigration load experiment may have driven the outcome of the resource use strategy. Indeed, in our study we found that high immigration rates resulted in a decrease in carrying capacity with adaptation (fig. 3; table S5). We found that individuals may become more resource efficient (fig. 2) under intermediate immigration levels, while low/high immigration levels point toward an increased use of resources (fig. 2). The above-mentioned hypothesis where higher intraspecific competition may evolve a strategy to increase resource use can be applied to high immigration rates as well. The main question is why the populations in the lowest level of immigration again become less resource efficient. A potential hypothesis could be related to the opportunities to evolve a resource efficient strategy. Under low levels of immigration, the amount of genetic variation that is available could be too low for evolution to act on (Lenormand 2002; Garant et al. 2007). Our findings may therefore represent a sampling effect: some individuals with higher fecundity due to a higher consumption rate were selected for, although they were not more efficient.

Haldane (1932) previously found that the selection of traits that increase performance can decrease carrying capacity. This implies that those individuals with a higher number of offspring will be selected for even though they consume more resources, leading to a decrease in the total population size. Such a negative effect of adaptation on population size has been found for the total yield of Saccharomyces cerevisiae (Jasmin et al. 2012) and was recently predicted by Abrams (2019), who used simple models to investigate how evolution influences population sizes. However, we have provided evidence in our study that this is not always the case: under certain circumstances selection for a more efficient use of resources may occur, leading to an increase in carrying capacity.

Our model also gave an indication of the time point at which the final adaptation started to influence the demography in our experiments. This was indicated by the reflection point of the sigmoid curve, which occurred earlier in the immigration load experiment than in the island biogeography experiment (about 25 and 75 days, respectively; table S3). We believe this to be due to the differences in immigration rates, as higher immigration rates in the immigration load experiment likely increased the chance for rare beneficial alleles to invade the population.

In summary, our study provides evidence for a coupling between the level of fecundity and the population growth factor and carrying capacity (evo-to-eco). Adaptation (i.e., higher fecundity) affected the population growth factor positively, while the influence on the carrying capacity was not univocal. A higher population growth factor can increase population stability and resilience and hence allow for evolutionary rescue, but it can also promote an increased risk for local population collapses because of overcompensatory regulations or enhanced spread, which is likely for species under scramble competition. Individuals producing more offspring more likely spread their genes in the population independent of whether this high fitness resulted from a higher resource use or from better resource efficiency. Here, we provide empirical evidence that certain ecological circumstances, such as differences in resource availability or immigration rates, can lead to the selection of higher resource efficiency or resource use, which could ultimately lead to higher or lower carrying capacity, respectively.

We thank Jelle van den Bergh for assisting during the research experiments, Giovanni Laudanno and Francisco Richter Mendoza for their modeling help, and Thomas Van Leeuwen for providing the strains of the stock population and the long-term adapted population of Tetranychus urticae. R.S.E. and K.B. thank the Netherlands Organization for Scientific Research (NWO) for financial support through a Vici grant (865.13.00) and a Nationale Wetenschapsagenda–Onderzoek op Routes door Consortia (NWA-ORC) grant (400.17.606/4175). K.B. thanks the Bijzonder Onderzoeksfonds (BOF) of Ghent University, and K.B. and A.A. thank the Ubbo Emmius sandwich program of the University of Groningen. D.B. and R.S.E. received funding from the Fonds voor Wetenschappelijk Onderzoek–Vlaanderen (FWO) research community “An eco-evolutionary network of biotic interactions” (W0.003.16N) and project G018017N. A.A. acknowledges the support of iDiv funded by the German Research Foundation (DFG FZT 118, 202548816), specifically funding through sDiv, the Synthesis Centre of iDiv.

K.B., A.A., D.B., and R.S.E. conceived the ideas and designed methodology; K.B. and A.A. collected the data; R.S.E. and K.B. designed the model and wrote the R code; K.B. analyzed the data; and K.B. led the writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication.

Data and code are accessible on DataverseNL at https://doi.org/10.34894/I0WXKL (Bisschop et al. 2021).

Literature Cited

“The lake-trout is one of the largest and most widely diffused of the Salmonidæ.” Figured: “Lake Trout (Salvelinus namaycush). Raquette Lake, New York.” From “Distribution and Some Characters of the Salmonidæ” by Tarleton H. Bean (The American Naturalist, 1888, 22:306–314).

Associate Editor: Chuan Yan

Editor: Daniel I. Bolnick