Abstract
With ongoing global change, life is continuously forced to move to novel areas, thereby imposing rapid changes in biotic communities and ecosystem functioning. As dispersal is central to range dynamics, factors promoting fast and distant dispersal are key to understanding and predicting range expansions. As the range expands, genetic variation is strongly depleted and genetic homogenisation increases. Such conditions should reduce evolutionary potential, but also impose severe kin competition. Although kin competition drives dispersal, we lack insights into its contribution to range expansions, relative to other causal processes. To separate evolutionary dynamics from kin competition, we combined simulation modelling and experimental range expansion using the spider mite Tetranychus urticae. Both modelling and experimental evolution demonstrated that plastic responses to kin structure increased range expansion speed by about 20%, while the effects of evolution and spatial sorting were marginal. This insight resolves an important paradox between the loss of genetic variation and earlier observed evolutionary dynamics facilitating range expansions. Kin competition may thus provide a social rescue mechanism in populations that are forced to keep up with fast climate change.
Introduction
Cooperation and conflict are central to understanding organismal interactions and their impact on population and community dynamics (West et al. 2002; Nadell et al. 2016). In group-living species, related individuals may profit from collective behaviour, but competition among kin may eventually outweigh the potential benefits (West et al. 2007). Dispersal provides a prominent means to avoid competition with kin and conflict more generally. Even when dispersal entails high costs, dispersers may be released from local competition, thereby increasing their inclusive fitness (Hamilton & May 1977). This effect is even stronger when dispersal is not a fixed genetic trait, but a conditional response (Clobert et al. 2009b; Bonte & Dahirel 2017), for instance conditional on kin structure where individuals plastically adapt their dispersal strategy to current levels of relatedness (Bitume et al. 2013). Dispersal has been recognized as a central and independent trait in life history, known to have a strong impact on spatial dynamics in fragmented landscapes or during range expansions (Kubisch et al. 2014; Bonte & Dahirel 2017; Cheptou et al. 2017; Legrand et al. 2017)(Kubisch et al. 2014; Bonte & Dahirel 2017; Cheptou et al. 2017; Legrand et al. 2017). Nonetheless, we surprisingly lack knowledge about the consequences of the interaction between kin structure and conditional dispersal for ecological patterns at large spatial scales, such as range expansions.
The speed and extent of range expansions and biological invasions have traditionally been regarded as consequences of human introductions or ecological factors such as enemy release (Keane & Smith 2002). Recently, a booming field of theory has demonstrated the importance of evolutionary dynamics through spatial selection of dispersal and/or reproduction at the expanding range front (Shine et al. 2011). The process of genetic assortment at expanding range borders results in the evolution of increased dispersal because highly dispersive genotypes colonize vacant habitat first. In addition, systematic low densities at the leading edge select for increased reproductive performance. Emerging assortative mating from spatial sorting then accelerates these evolutionary dynamics at the range front (i.e. the Olympic village effect (Phillips et al. 2010)), speed up range expansions and make biological invasions even more challenging to contain (Phillips 2015). Although mechanisms behind spatial selection and spatial sorting are different, we refer here to both as spatial sorting for the ease of communication.
Paradoxically, evolutionary change should be slow during range expansions as genetic variation already gets depleted early-on due to subsequent founder effects, rendering drift important, and thus potentially constraining further evolutionary change (Weiss-Lehman et al. 2017). Simultaneously, high levels of local genetic relatedness emerge due to reduced population sizes (Newman & Pilson 1997; Kubisch et al. 2013; Nadell et al. 2016). In many arthropods, for instance, single female colonisers found highly related populations (Dingle 1978). Kin competition in combination with an appropriate conditional dispersal response may thus be a key driver of fast range expansions and could potentially explain the paradox of fast expansions despite severe genetic diversity loss (Estoup et al. 2016). Unfortunately, conditional dispersal related to kin interactions have to date been neglected in the context of range expansions and biological invasions (but see one study on the evolution of unconditional dispersal and kin structure during range expansions (Kubisch et al. 2013)).
We tested the relative effect of kin competition and spatial sorting on range expansion dynamics by means of in silico simulations and experimental range expansions, using the two-spotted spider mite Tetranychus urticae Koch (Acari, Tetranychidae) as a model organism. This mite species allows us to assess quantitative life history traits in detail (Macke et al. 2011; Fronhofer et al. 2014; De Roissart et al. 2016; Van Petegem et al. 2016). Spider mite life history traits, including dispersal, are documented to be heritable but highly plastic in response to inter- and intra-generational environmental and social conditions (Magalhães et al. 2009; Turcotte et al. 2011; Ochocki et al. 2017). The effect of genetic relatedness on both dispersal distance and emigration rate is, for instance, as strong as that of density dependence (Bitume et al. 2014).
Material and Methods
General model algorithm
The model is individual-based and object-oriented and simulates demographic and evolutionary processes along a one-dimensional array of patches. Patches contain resources, which are consumed by individuals at different rates depending on their life stage (juvenile or adult). Resources are refreshed weekly. Individuals start out with a limited reserve of resources, which they will need to replenish in order to survive. A detailed model description and additional results on in silico trait evolution are available in SUPPLEMENTARY MATERIAL 1.
Males and females of Tetranychus urticae differ in a number of aspects. Firstly, males are much smaller when reaching the adult life stage, and hence contribute little to resource consumption (males hardly feed when becoming adult). Secondly, dispersal behaviour differs between the two sexes, with adult females being the dominant dispersers, whereas juveniles and males disperse very little. Lastly, the species is characterized by a haplodiploid life cycle, where non-mated females only produce haploid male offspring, and mated females can produce both haploid male and diploid female eggs. The sex ratio of spider mites is female-biased, with approximately 0.66 males to females. For these reasons and for the sake of simplicity, we designed the model to only include female mites, where the genotype of the individual is passed on from mother to daughter. Individuals carry the following genetic traits: age at maturity, fecundity, longevity, and a categorical neutral genotype (one unique allele per individual) which defines relatedness. Mean relatedness of an individual A in a patch X can be calculated as the number of individuals in patch X carrying the same relatedness genotype as individual A, divided by the total number of individuals, and hence ranges from 0 (no related individuals present) to 1 (all individuals are related to individual A). After 80 steps, concurring with 80 days in our experimental range expansions, both the length of the metapopulation (i.e. the total number of patches in the metapopulation) and the mean life history trait values at the core and edge were recorded. To this end, individuals present in the first patch of the metapopulation (core) or in the last three occupied patches (edge) were tracked (cf. the experimental part of the study) and the mean value of every life history trait was calculated and recorded.
The following scenarios were tested:
A treatment where dynamics include putative kin competition and evolution. In this scenario, females pass their allele values to the offspring. Mutations occur at a rate of 0.001 and change the trait value to a randomly assigned one as during the initialisation phase. The genotype ID remains unchanged (relatedness is unaffected by trait value mutation).
A treatment where dynamics do not allow evolution, by changing trait values during reproduction at a mutational rate of 1. In this scenario, all trait values are reset according to the initialisation procedure. Only genotype is maintained, and therefore kin structure and possible kin competition.
A treatment representing a reshuffling of females. Under this scenario, and as in the experimental procedure, adult females are replaced each week by random females from the stock population. Thus, both trait values and relatedness genotype are re-initialised, eliminating both kin competition and evolution.
The model was entirely programmed in Python 3.3. Syntax of the code is publically available on Github: git@github.ugent.be:dbonte/Python-code-Van-Petegem-et-al.-2017.git
Experimental range expansions
T. urticae strains
Several different strains of T. urticae were used within the current study: LS-VL, MR-VP, SR-VP, JPS, LONDON, and MIX. The LS-VL strain was originally collected in 2000 on rose plants in Ghent (Belgium) and since then maintained on common bean (Phaseolus vulgaris, variety Prélude) in a laboratory environment (Van Leeuwen et al. 2004). This strain is known to harbour sufficient genetic diversity for studies of experimental evolution (Van Leeuwen et al. 2008; De Roissart et al. 2016). The MR-VP, SR-VP, JPS, ALBINO and LONDON strains, in contrast, were collected from different greenhouses and completely devoid of any genetic variation by consistently inbreeding mothers with sons over seven generations (see Díaz-Riquelme et al. 2016 for the followed procedure). By crossing these five different isofemale strains, a strain containing substantial genetic variation was created: the MIX strain. This was done by reciprocally crossing males and females of each of the isofemale strains: for each combination of strains, one female (last moulting stage) of strain X/Y was put together on a bean patch with three males of strain Y/X, allowing fertilisation (in case a fertilisation was unsuccessful, this step was repeated). The resulting F1, F2, and F3 generations were again mixed in such a manner that, eventually, we obtained one mixed strain (MIX) that comprised a mixture of all isofemale strains. Stock populations of the LS-VL and MIX strain were maintained on whole common bean plants in a climate-controlled room (28.1°C ± 2.1°C) with a light regime of 16:8 LD, while stock populations of the ISO strains were maintained on bean leaf rectangles in separate, isolated incubators (28°C, 16:8 LD). Before eventually using the mite strains to initialise the experimental metapopulations, they were first synchronised. For each strain, sixty adult females were collected from the respective stock populations, placed individually on a leaf rectangle of 3.5 × 4.5 cm2, and put in an incubator (30°C, 16:8 LD). The females were subsequently allowed to lay eggs during 24 hours, after which they were removed and their eggs were left to develop. Freshly mated females that has reached the adult stage on day prior to mating of the F1 generation were then used to initialise the experimental metapopulations (see below). As all mites were kept under common conditions during this one generation of synchronisation, direct environmental and maternally-induced environmental effects (Macke et al. 2011) of the stock-conditions were removed.
Experimental range expansion
An experimental range expansion consisted of a linear system of populations: bean leaf squares (2 × 2 cm2) connected by parafilm bridges [8×1 cm2], placed on top of moist cotton. A metapopulation was initialised by placing ten freshly mated one-day-old adult females on the first patch (population) of this system. At this point, the metapopulation comprised only four patches. The initial population of ten females was subsequently left to settle, grow, and progressively colonise the next patch(es) in line through ambulatory dispersal. Three times a week, all patches were checked and one/two new patches were added to the system when mites had reached the second-to-last/last patch. Mites were therefore not hindered in their dispersal attempts, allowing a continuous expansion of the range. A regular food supply was secured for all populations by renewing all leaf squares in the metapopulation once every week; all one week old leaf squares were shifted aside, replacing the two-week-old squares that were put there the week before, and in their turn replaced by fresh patches. As the old patches slightly overlapped the new, mites could freely move to these new patches. Mites were left in this experimental metapopulation for approximately ten generations (80 days) during which they could gradually expand their range.
Treatments
We performed two experiments, in each of which experimental metapopulations were each time assigned to one out of two different treatments. In the first experiment, they were assigned to either “NMP” or “RFS”. In the NMP treatment (referring to “non-manipulated population”), experimental metapopulations were initialised using mites from the LS-VL strain. The metapopulations within this treatment thus started with a high enough amount of standing genetic variation for evolution to act on. Kin structure was not manipulated in this treatment and kin competition was therefore expected to increase towards the range edge (see introduction). In the RFS treatment (standing for “Replacement From Stock”), experimental metapopulations were also initialised with mites from the LS-VL strain, but all adult females in the metapopulations were replaced on a weekly basis by randomly chosen, but similarly aged, females from the LS-VL stock. As a result, any spatial sorting of phenotypes was nullified and kin structure randomized and hence no longer expected to increase towards the range edge. The spatial structure (local densities) of the metapopulations within this treatment was however maintained (i.e., if x females were on a patch before the replacement, they were replaced by x females from the stock). In this first experiment, we thus compared unmanipulated, genetically diverse metapopulations (NMP treatment) with regularly reshuffled metapopulations where only effects of density-dependent dispersal remained (RFS treatment) (cf. (Ochocki et al. 2017)). Both treatments were replicated six times.
In the second experiment, experimental metapopulations were assigned to either “MIX” or “ISO”. In the MIX treatment (for “mixture of inbred lines” –see above), experimental metapopulations were initialised using mites from the MIX strain. This strain harboured standing genetic variation on which evolution could act. No manipulations of kin structure were performed. In the ISO treatment (for “ISOfemale line”), experimental metapopulations were initialised using mites from the SR-VP, JPS or LONDON isofemale strain (originally, there were also metapopulations for the MR-VP and ALBINO strain, but these experimental metapopulations collapsed very early within the experiment). These metapopulations therefore harboured no standing genetic variation for evolution to act on. As in the MIX treatment, kin structure was not manipulated. In this second experiment, we thus compared unmanipulated metapopulations (MIX treatment) with metapopulations where only condition dependency (density-dependent dispersal and kin competition) played a role (ISO treatment) [cf.(Wagner et al. 2017)]. Both treatments were replicated six times (in case of ISO, two replicates (i.e., experimental range expansions) per isofemale strain were set up).
In addition to monitoring range expansions along the linear system, we quantified life history trait variation genetic variation in gene expression between core and edge populations (details in SUPPLEMENTARY MATERIAL 3, 4)
Results
First, we formalized our hypotheses by means of a highly parameterized, but simple simulation model based on spider mite life histories and relatedness-dependent dispersal reaction norms. We simulated one-dimensional range expansion during over 8-10 generations (80 days, SUPPLEMENTARY MATERIAL 1). Despite the incorporation of uncertainties regarding condition-dependent dispersal thresholds, the model predicted range expansion to proceed at a 25.9% slower mean rate when signatures of both kin competition and spatial sorting were removed, while expansion rates were only 7.4% slower when spatial sorting was prevented, but kin competition was present (Fig. 1). Thus, range expansion speed increased in our model with 21% by kin competition, but only by 1% due to spatial sorting
Second, to test this prediction we ran two parallel experiments where we started experimental range expansions with a limited amount of founders (10 females), thereby mimicking ongoing range expansion of T. urticae along a linear patchy landscape. Each replicated population invaded its respective landscape for ten generations (spanning 80 days). Two parallel experiments were conducted in which genetic diversity and relatedness were manipulated to infer the relative importance of spatial sorting and kin competition for range expansion dynamics. To specifically test for evolution during range expansion, we determined quantitative genetic differences in life-history traits between core and edge populations and measured patterns of gene expression.
In a first experiment, we compared the dynamics of range expansion and relevant life history traits between six replicated non-manipulated experimental range expansions (NMP) and six replicates where single adult females were randomly replaced by similar-aged females from the same stock population (further abbreviated as RFS - “Replacement From Stock”). The latter treatment maintains age and population structure but prevents genetic sorting, and destroys local relatedness, thus preventing both spatial sorting and kin competition. In a second parallel experiment, six replicated experimental range expansions with either depleted or substantial standing genetic variation were contrasted (Turcotte et al. 2011; Ochocki et al. 2017; Wagner et al. 2017). Single, different isofemale lines were used for the experiments with strongly depleted genetic variation (further abbreviated as ISO), and recombined isofemale lines were used for the genetically enriched experimental expansions (further abbreviated as MIX). The enriched populations maintain density, genetic, and phenotypic structure. In the genetically depleted lines, spatial sorting is impossible but relatedness patterns are left intact. Quantitative genetic trait variation as determined in common garden experiments did not differ among any of the lines, likely due to the dominance of plasticity [see SUPPLEMENTARY MATERIAL 2]. Starting from the same levels of trait variation, MIX and NMP thus represent treatments where range expansions are determined by spatial sorting and putative kin interactions, ISO represents the treatment where kin structure is high but where spatial sorting is restricted, and RFS a treatment with both kin competition and spatial sorting constrained.
For each replicate, we measured range expansion dynamics and genotypic trait structure at an unprecedented level of detail. By counting the number of adult females three times per week on each of the occupied patches during the experimental range expansion, we detected a 28% lower rate of range expansion in the RFS scenario, in which kin competition and spatial sorting were constrained, versus the unconstrained NMP scenario (day × treatment interaction, F1,54.8=7.62; P=0.007; Fig. 3). However, no statistically significant differences were found between the unconstrained MIX treatment and the ISO treatment, which inhibited spatial sorting but left kin competition intact (day × treatment interaction, F1,71.1=0.71; P=0.40). Differences in slopes were 0.082 ± 0.026 SE patches/day for the NMP-RFS comparison and 0.030 ± 0.036 SE patches/day for the MIX-ISO comparison, with eventual range size matching the average model predictions. In contrast to other studies finding evolution leading to either increased (Ochocki et al. 2017; Weiss-Lehman et al. 2017) or reduced variation among replicates (Williams et al. 2016b; Wagner et al. 2017) no significant differences in the variation in spread rate were present among any of the treatments (coefficients of variation in the reached distance with 95% CI based on bootstrapping: NMP: 0.246 [0.147-0.276]; RFS: 0.2424 [0.133-0.279]; MIX: 0.279 [0.207- 0.314]; INBRED: 0.220 [0.198-0.248]).
We subsequently tested whether increased range expansion resulted from evolved trait differences between edge and core populations [see SUPPLEMENTARY MATERIAL 3]. No significantly higher dispersal rates were detected in individuals from the expanding front, relative to the ones collected from the core patches. Therefore, the accelerated range advance in the treatments with unconstrained evolutionary dynamics was achieved independently of evolutionary changes in dispersiveness. Consistent with predictions of enhanced intrinsic growth rates leading to faster range expansions and sorting of traits at the expansion front (Burton et al. 2010), we found evolved differentiation in the intrinsic growth rate. Intrinsic growth rates were systematically higher in edge relative to core populations in experiments that allowed evolutionary dynamics (NMP: F1,153=5.32, p=0.0225; MIX: F1,235=6.46, p=0.0117; See Fig. 4), but not in those where evolution was experimentally inhibited.
Discussion
The destruction of spatial genetic structure, and thus of both kin competition and spatial sorting, resulted in a lower expansion rate than just the inhibition of spatial sorting by depleted genetic variation. In all treatments except RFS, kin competition was high due to serial founder effects following small population sizes at the initiation of the experiments. Spatial selection was not pronounced in the experimental range expansions that allowed evolution. The impact of evolved differences in growth rates on range expansions were therefore only marginal relative to the impact of eliminating the potential for kin competition. The stronger inhibition of range expansion in the RFS treatment did not result from lower levels of trait variation and can therefore only be attributed to the elimination of kin competition, and not by different evolutionary trajectories. Such phenotypic variation despite genetic depletion is probably widespread in wild populations and typically maintained by individual differences in development (Cressler et al. 2017), but we can neither exclude long term intergenerational plasticity (Bitume et al. 2014).
Evolved increased growth rates at the leading expansion front accord with processes of spatial selection at the expanding front and are in line with recent studies based on field observations (Phillips et al. 2010; Shine et al. 2011; Alex Perkins et al. 2013; Van Petegem et al. 2016). Systematically low densities at the range front select for increased reproduction, and such strategies are known to advance range expansion substantially (Phillips et al. 2010; Shine et al. 2011; Van Petegem et al. 2016). Surprisingly, we found no indications of variation in any single fitness-related trait among the different treatments. We observed significant replica*location variation in traits during experimental evolution, eventually resulting in divergent trait covariances among replicates (see SUPPLEMENTARY MATERIAL 3). Under such conditions, different life history strategies encompassing multivariate trait correlations but leading to similar high population growth rates might eventually be spatially sorted. Bootstrapping of the vital rates within replicates could however, not confirm the empirically determined higher growth rates at the leading edges (see SUPPLEMENTARY MATERIAL 3). We therefore attribute this opposing evidence from simulated relative to observed intrinsic growth rates to the fact simulated one systematically assume invariant life trait expression during population growth, so neglecting density effects and other individual interactions. Because the interpretation of fitness should be tested under relevant, and varying realistic demographic conditions (Bonte et al. 2014), our empirical assessment thus approaches realistic conditions better than theoretically composed ones.
We here can exclude heterosis as a driving mechanism leading to a “catapult effect” and subsequent faster range expansions in the MIX treatment (Wagner et al. 2017) since metapopulations were initialised with an already mixed strain instead of with separate strains that would hybridise after initialization. We neither did find differences in quantitative genetic (SUPPLEMENTARY MATERIAL 3) and transcriptomic (SUPPLEMENTARY MATERIAL 4) trait variation between the inbred, mixed, and highly diverse stock population, or between core and edge populations, again indicating the dominance of plasticity over evolution for life history trait expression in our model system. Ambulatory dispersal in mites has a genetic basis (Yano & Takafuji 2002) but is simultaneously highly dependent on differences in density, also those experienced in earlier generations by parents and grandparents (Bitume et al. 2014). We assessed the mites’ dispersal behaviour under conditions that reflect the low-density conditions in (Bitume et al. 2014) for which grandparental environmental conditions did not affect dispersal behaviour. The accelerated range expansion does not result from elevated densities at the front, and thus density dependency in dispersal, neither did we find evolution of increased competitive abilities through enhanced foraging efficiency or increased long-distance dispersal at the range front (SUPPLEMENTARY MATERIAL 3).
With exception of the RFS treatment, kin competition is expected to be high due to serial founder effects following small population sizes at the initiation of all experiments. Kin competition causes more (Bowler & Benton 2005; Clobert et al. 2009a) and further dispersal (Bitume et al. 2013), and is hence predicted to speed-up range expansion. Kin recognition is not restricted to animals, and has been demonstrated in several plant species as well (Dudley & File 2007; Murphy & Dudley 2009; Biedrzycki et al. 2010; Dudley et al. 2013; Crepy & Casal 2015). Evidence for kin recognition is specifically accumulating in Arabidopsis (Biedrzycki et al. 2010), the model system used for experimental range expansions by Williams et al. (Williams et al. 2016a)). They performed a reshuffling experiment and did not consider the obliteration of kin interactions with such experimental procedure. Because kin recognition is widespread and not restricted to animals, it may be a highly important but neglected driver of range expansions for a wide variety of life forms. In contrast to Williams et al’s work where sorting narrowed variance in spread rate and parallel but independent experiments (Ochocki et al. 2017; Weiss-Lehman et al. 2017) that reports higher spread rate variance, we did not detect any changes in spread rate variation. No general conclusions on the impact of spatial sorting on spread predictability can thus be made as they will likely depend on many joint ecological, evolutionary and social factors.
The obliteration of spatial genetic structure, and thus both kin competition and spatial sorting, resulted consequently in a lower expansion rate relative to experiments were only spatial sorting was prohibited by depleting genetic variation. Our results provide the first evidence of kin competition as an overlooked but quantitatively highly significant driver of range expansions compared to spatial sorting. Emerging genetic structure and high relatedness per se along a range expansion front can thus be responsible for fast range expansions, even in the absence of substantial sorting of individual life history traits. The loss of genetic variation during range expansions and biological invasions is typically considered to be a limiting factor because it constraints the potential for local adaptation. We here show that, on the contrary, that it may actually lead to faster range expansions, impose social rescue and therefore allow population to keep pace with high rates of climate change.
Acknowledgements
This project was funded by the Fund for Scientific Research–Flanders (FWO; project G.0610.11 and G.018017N). D.B. and R.S. were supported by the FWO research network EVENET; RS by the KU Leuven Excellence Center Financing PF/2010/07. Services used in this work were provided by the Flemish Supercomputer Center (VSC), funded by Ghent University, the Hercules Foundation, and the Flemish government (Department of Economy, Science, and Innovation).
Appendices
Supplementary material 1: Details of the simulation model
Supplementary material 2: Quantitative trait variation at the onset of the experimental evolution
Supplementary material 3: Contrasting quantitative trait (co)variation between core and edge populations
Supplementary material 4: Methods and results transcriptomics