The NCEAS/NESCent Working Group on Genetic Monitoring (GeM) is chaired by F.W. Allendorf, University of Montana, USA, and M.K. Schwartz, USDA Forest Service, USA. The other members are: C.S. Baker (Oregon State University, USA), D.P. Gregovich (University of Alaska, USA), M.M. Hansen (Aarhus University, Denmark), J.A. Jackson (Oregon State University, USA), K.C. Kendall (US Geological Survey, USA), L. Laikre (Stockholm University, Sweden), K. McKelvey (USDA Forest Service, USA), M.C. Neel (University of Maryland, USA), I. Olivieri (Université de Montpellier II, France), N. Ryman (Stockholm University, Sweden), R. Short Bull (University of Montana, USA), J.B. Stetz (University of Montana, USA), D.A. Tallmon (University of Alaska Southeast, USA), C.D. Vojta (USDA Forest Service, USA), D.M. Waller (University of Wisconsin, USA) and R.S. Waples (National Marine Fisheries Service, USA)
Widespread environmental changes including climate change, selective harvesting and landscape alterations now greatly affect selection regimes for most organisms. How animals and plants can adapt to these altered environments via contemporary evolution is thus of strong interest. We discuss how to use genetic monitoring to study adaptive responses via repeated analysis of the same populations over time, distinguishing between phenotypic and molecular genetics approaches. After describing monitoring designs, we develop explicit criteria for demonstrating adaptive responses, which include testing for selection and establishing clear links between genetic and environmental change. We then review a few exemplary studies that explore adaptive responses to climate change in Drosophila, selective responses to hunting and fishing, and contemporary evolution in Daphnia using resurrected resting eggs. We further review a broader set of 44 studies to assess how well they meet the proposed criteria, and conclude that only 23% fulfill all criteria. Approximately half (43%) of these studies failed to rule out the alternative hypothesis of replacement by a different, better-adapted population. Likewise, 34% of the studies based on phenotypic variation did not test for selection as opposed to drift. These shortcomings can be addressed via improved experimental designs and statistical testing. We foresee monitoring of adaptive responses as a future valuable tool in conservation biology, for identifying populations unable to evolve at sufficiently high rates and for identifying possible donor populations for genetic rescue. Technological advances will further augment the realization of this potential, especially next-generation sequencing technologies that allow for monitoring at the level of whole genomes.
Species across the globe are experiencing drastic changes in environmental conditions as a result of human activities. Climate change induced by increased greenhouse gas emissions has emerged as a top concern, both scientifically and politically (Kerr 2007; Moss et al. 2010), with increasing evidence that it has already affected populations of many plant and animal species (Parmesan 2006). Organisms also confront other environmental challenges like pollution (Anderson et al. 1994), emerging pathogens (Parker & Gilbert 2004), and newly introduced species that alter patterns of predation or competition (Mack et al. 2000). Finally, the intensified harvesting of wild populations via fishing and hunting radically alters population demographics and selection regimes (Allendorf et al. 2008; Coltman 2008).
In the short term, animals and plants acclimate to shifting environmental conditions via phenotypic plasticity, developing and expressing particular traits in response to local environmental conditions. Over the longer term, in a second kind of response, organisms may disperse to more favorable sites, potentially over long distances. A third type of response is via genic selection leading to adaptation. It is important to be able to differentiate among these types of responses and to consider their relationships. Of note, some organisms cannot disperse to more favorable conditions because of low vagility or habitat fragmentation in human-altered environments. For this reason, their ability to acclimate or adapt locally becomes vital and is all the more important because fragmentation itself selects against dispersal propensity (Ronce 2007). A firm distinction must also be made between acclimation resulting from phenotypic plasticity, which allows for response to environmental change in the short term, and adaptive evolutionary change, which allows for adaptation to a continually changing environment beyond the limits of possible responses attributable to phenotypic plasticity (Gienapp et al. 2008); Box 1 addresses this issue in more detail.
Table Box 1. Types of response – acclimation or adaptation to environmental change?
Phenotypic traits appear to be changing faster in populations subject to obvious anthropogenic forces than in those persisting in more natural habitats (Hendry et al. 2008). Inferring adaptation from shifts in phenotype, however, runs the risk of confusing short-term, plastic responses with longer-term genetic changes (Gienapp et al. 2008; Hendry et al. 2008). Most individuals have the capacity to respond to local environmental conditions via phenotypic plasticity. If phenotypic changes mostly reflect plasticity, responses to continuing environmental change will be limited (Gienapp et al. 2008). However, plasticity can also serve to provide an initial rapid response to environmental change that can then facilitate subsequent genetic adaptation via ‘genetic assimilation’ (Lande 2009). In addition, when environmental stresses reverse, as could be the case when policies to harvest wild populations change, phenotypic changes reflecting plasticity would also reverse rapidly, in contrast to genetic changes, which would be harder to reverse.
To infer actual adaptive change and to forecast longer-term responses to environmental change, we need to rigorously distinguish genetic adaptation from plasticity. This distinction can be a complex task because most phenotypic traits are polygenic and genes can interact epistatically. Dissecting the genetic basis for quantitative traits usually requires experiments such as common garden or reciprocal transplant setups (Endler 1986; Lynch & Walsh 1998). These classic approaches decompose total variation into its components, namely additive genetic, non-additive genetic, environmental, and genetic-by-environment (G × E) interactions (Lynch & Walsh 1998).
Plasticity reflects shifts in phenotype due to non-genetic responses to variation in environmental conditions. However, because these reaction norms themselves have a genetic basis, they can also evolve (Pigliucci & Schlichting 1998; Czesak et al. 2006). It would therefore be interesting to monitor changes in these reaction norms using, e.g. resurrection techniques involving diapausing seeds or eggs from different time periods (Cousyn et al. 2001; Franks et al. 2007).
As the preceding makes evident, reasons are numerous for studying adaptive responses to environmental change in wild populations. In a general scientific context, human-disturbed environments represent a series of unplanned, large-scale, ‘natural’ experiments for observing evolution in action. From an applied perspective, we are interested in (i) evaluating and forecasting if and how populations and species can adapt to and persist in human-altered environments; (ii) assessing if individual populations can invoke an adaptive response and, if not, take measures such as genetic rescue (Richards 2000; Ingvarsson 2001) to assist adaptive processes; (iii) assessing and mitigating unintended consequences of anthropogenic change, such as selection imposed by harvesting of wild populations and (iv) using evidence for adaptive responses as an indicator of environmental stress that warns of incipient population declines. Concerning the latter point, it is important to note that even if adaptive responses occur, extinction of populations may still result. An increasing ‘lag load’ may develop as environmental changes accumulate and accelerate, whereas at the same time, the mean phenotype of the population increasingly lags behind the optimal phenotype, leading to an increasing burden of selective deaths (Box 2).
Table Box 2. Lag load and rates of evolution
Following Haldane (1949), evolutionary biologists characterize evolutionary change as a change in mean trait value relative to the variation we observe around that mean wherein one Haldane represents the change in mean per generation relative to its phenotypic standard deviation (for details, see Hendry & Kinnison 1999). Phenotypic means can also change because of phenotypic plasticity, so this measure refers only to changes resulting from genic selection. Haldanes are measured relative to generation time, which necessitates adjustment of this quantity to compare rates of adaptation among organisms and to rates of environmental change. In particular, to measure absolute rates of evolutionary change (RoE) in units of phenotypic standard deviations, we need to adjust by generation time as follows:
This unit of measure allows comparison of the rates at which traits evolve among populations and species.
Such evolutionary responses are key for understanding how well populations can track changes in their environment. Felsenstein (1971) and Maynard Smith (1978) first developed the idea of lag load to refer to the situation in which continuous changes in the environment produce a gap between the optimum trait value and the mean genotypic value for that trait actually observed in a population. The size of this lag load depends on the rate of environmental change relative to the rate of adaptation. As environmental change accelerates and this lag load increases, the intensity of selection will also increase. The amount of selection that can occur, however, is limited by both the reproductive excess present in the population and the heritability (h2) of fitness in these new environments. Each of these will tend to decline as selection forces intensify and diversify and as populations depart further from their optima. Theory suggests that most populations’ maximum sustainable rate of evolution is considerably less than 0.1 Haldanes per generation (Lynch & Lande 1993; Burger & Lynch 1995).
Tracking adaptive responses in populations requires monitoring, i.e. sampling and analysing populations over time. In ecology, a common practice is to trade time for space, and studying phenotypic traits or molecular markers within the same species across environmental gradients can be used for forecasting adaptive responses to environmental change in local populations (e.g. Byars et al. 2007; Jensen et al. 2008; Riba et al. 2009). Nevertheless, in many cases, populations have experienced different evolutionary histories and differ in important demographic characteristics such as effective population size and gene flow and in adaptation to local environments. Populations may therefore from the onset differ in standing genetic variation of potential adaptive value under environmental change. Furthermore, empirical evidence has shown that parallel phenotypic change in different populations can reflect both similar and different genetic architecture of traits. For instance, threespine sticklebacks (Gasterosteus aculeatus) from marine populations have repeatedly colonized freshwater habitats, and extensive parallel evolution of armor loss has occurred. In all cases examined, this adaptive response has involved variation at the Ectodysplasin (Eda) locus (Colosimo et al. 2005). In sharp contrast, different genetic architecture of the same phenotypic trait underlies parallel evolution of adaptive colour variation in rock pocket mice (Chaetodipus intermedius) (Nachman et al. 2003) and beach mice (Peromyscus polionotus) populations (Hoekstra et al. 2006). We therefore argue that adaptive responses to environmental change are best studied by observing the process using a temporal approach, i.e. genetic monitoring, rather than predicting responses based on the distribution of adaptive variation across populations in different environments. Of interest, one of the classic textbook examples of contemporary evolution, industrial melanism in the peppered moth (Biston betularia), can be considered an example of genetic monitoring of adaptive change (Bateson 1900; Kettlewell 1961).
In this review, we explore how to monitor the genetic composition of populations to infer whether adaptive changes are occurring. Schwartz et al. (2007) broadly defined genetic monitoring as the tracking of neutral genetic markers through time to estimate demographic and/or population genetic parameters. Here, we focus on studies using temporal genetic data to identify adaptive genetic changes and how such changes can be related to changing environments. We consider analyses of both molecular markers [e.g. candidate genes, quantitative trait loci (QTLs), or genome scans] and phenotypic traits. Our overarching aim is to provide a practical framework for identifying adaptive evolutionary responses to environmental change using genetic monitoring. We first describe and discuss the use of molecular markers and phenotypic traits in genetic monitoring. Next, we develop specific criteria for demonstrating adaptive evolutionary responses to specific shifts in environmental conditions. In this case, we are inspired by the approach by Endler (1986) for developing criteria for inferring natural selection in the wild. We then search the literature for and identify examples of genetic monitoring of adaptive change to illustrate the various possible approaches. This groundwork leads into a review of a larger body of studies, again with reference to how well they meet the criteria for demonstrating adaptive responses. Finally, we discuss the conservation applications and future perspectives of monitoring adaptive change in populations.
Molecular markers and phenotypic traits for genetic monitoring of adaptive change
Two targets exist for monitoring adaptive change: molecular markers or phenotypic traits, the latter encompassing qualitative traits preferably exhibiting simple Mendelian inheritance or quantitative traits. Each approach has its advantages and disadvantages. For example, a major challenge in monitoring quantitative traits is distinguishing between phenotypic plasticity and adaptive genetic responses (Box 1). Monitoring molecular markers avoids this problem but raises another, namely how to associate genetic variation at particular loci with variation in phenotypic traits and the selective agent acting on that variation (Vasemägi & Primmer 2005).
It will almost always be necessary to evaluate genetic change at candidate loci under possible selection against a background of several presumably neutral loci, such as microsatellites or single nucleotide polymorphisms (SNPs). The number of loci needed for background will depend on the power of the specific test applied, but as a general recommendation derived from one of the most commonly applied statistical tests, it should exceed 20 (Beaumont & Nichols 1996).
Because most adaptive traits have a polygenic genetic architecture, failure to demonstrate genetic change at a specific candidate locus does not rule out that change has occurred at other loci affecting the same trait. This complexity argues for analysing many loci (hundreds to thousands) to densely cover the genome and possibly identify other loci under selection from the same environmental stressors. Fortunately, this approach is now becoming feasible with the advent of next-generation sequencing methods (Margulies et al. 2005). Transcriptome sequencing provides rich sources of SNPs (Barbazuk et al. 2007), facilitating identification of the genes involved in adaptive change (e.g. Renaut et al. 2010; Williams & Oleksiak 2011). An exciting recent development is the use of reduced-representation genome-wide sequencing; thousands of short fragments are sequenced throughout the genome, allowing for genotyping by sequencing of thousands of SNPs and narrowing the gap between model and non-model species (Allendorf et al. 2010; Davey et al. 2011).
We are often limited in our ability to collect samples on a sufficiently long temporal scale, particularly for species with long generation times. However, we can use ‘‘retrospective sampling’’ to extend the time scale using either samples or data from previous studies. For instance, Umina et al. (2005) and Balanya et al. (2006) analysed adaptive responses to climate change in Drosophila species based on ca. 25-year-old historical data. The disadvantage of this approach is that molecular analyses of contemporary samples have to be based on the same approaches as those used in the historical studies; in the case of Drosophila, these methods involved analysis of allozyme and chromosome polymorphisms. If historical samples are available, such as herbarium specimens of plants or hard parts from animals (e.g. bones, feathers, or fish scales), they can be used as sources of DNA. Wandeler et al. (2007), Leonard (2008) and Nielsen & Hansen (2008) have reviewed these approaches along with the inherent challenges of DNA degradation and potential contamination. It should be noted that current protocols for reduced-representation genome-wide sequencing require high amounts of high-quality DNA (Davey et al. 2011), thus limiting their applicability for historical DNA samples.
Testing for selection at molecular markers
Genetic monitoring of adaptive change at the molecular level involves testing for selection on ecological time scales. This calls for the use of outlier test approaches in which loci under putative selection (either hitch-hiking or direct selection) are identified as outliers in terms of genetic differentiation and/or reduced variation (selective sweep) against a background of supposedly neutral markers (Storz 2005). Although we have not found any empirical examples, it should in principle also be possible to test for association between allelic variation and environmental parameters that change over time, using some of the available landscape genomics and association tests, notably that described by Coop et al. (2010). In Table 1, we list tests of relevance for monitoring, along with their requirements and potential drawbacks, and we elaborate on some of the methods below.
Table 1. Tests and approaches for genetic monitoring of adaptive changes using molecular data
Potential problems for genetic monitoring
FST-based outlier tests
Detection of loci that show significantly high (or low) differentiation. Significance determined by simulations assuming specific population structure models
Co-dominant (in some cases also dominant) markers. Many loci, >20. Temporal samples, increasing statistical power with number of samples
Geographical demographic models are assumed, whereas monitoring involves temporal sampling from the same populations
Selective sweeps lead to reduced variation. Comparison of the ratio R of Θ = 4Neμ for each locus between a pair of populations (Ne = effective population size, μ = mutation rate). Θ is estimated based on V (variance in repeat number at microsatellite loci) or H (expected heterozygosity). Outliers detected empirically, assuming normal distribution of ln(RH) or ln(RV). A Bayesian test based on ln(RV) has been developed that allows for analysing multiple populations (Marshall & Weiss 2006)
Co-dominant markers, lnRV applicable only to microsatellite loci. Many loci (preferably >20), as significance is determined from empirical distributions. Temporal samples, pairwise tests (or tests involving multiple samples using the method by Marshall & Weiss (2006))
Different phases of selective sweeps may generate patterns other than decreased variation. For instance, positive selection for a rare allele at a bi-allelic locus will initially increase heterozygosity
Testing for correlation between allelic variation and environmental parameters while controlling for neutral population structure
Temporal data on relevant environmental parameters. SNPs or other bi-allelic markers, though multi-allelic markers also useful for the method by Pritchard et al. (2000). Several (>>2) temporal samples required, depending on specific test
Assumes geographical structure, but should be applicable to temporal data
No generalized framework. Test procedures should verify that (i) clinal patterns reflect adaptive responses to environmental gradients; (ii) clines have changed over time and (iii) this corresponds to a change in environmental gradients
Temporal data on relevant environmental parameters and markers showing clinal patterns
Research possibilities restricted to cases where clines occur
Analysing large numbers of mapped markers, obtained either by linkage mapping, reduced-representation genome-wide sequencing and alignment to a reference genome (Davey et al. 2011), or analysing whole genomes, holds many advantages because it allows for fine-scale mapping by identifying signatures of selection along a chromosomal region (Nielsen 2005; Wiehe et al. 2007). Tests relying on signals from several linked markers reduce the number of false positives and also allow for more exact identification of the chromosomal region under selection (Nielsen 2005; Wiehe et al. 2007; Mäkinen et al. 2008). There is as yet very little reported experience with using genome-wide data for genetic monitoring of adaptive change in eukaryotes. However, Burke et al. (2010) used this type of approach to investigate the genomic footprints of selection in lines of Drosophila melanogaster selected for accelerated developmental time over 600 generations. Using next-generation sequencing of pooled samples, they identified several genomic regions with increased divergence between selected and control lines, presumably reflecting selection, but no regions where classical selective sweeps had led to fixation of alleles.
If clinal variation at molecular markers reflects selection, then monitoring clinal patterns can provide evidence for adaptive responses to environmental change, such as climate change (Umina et al. 2005; Balanya et al. 2006). It should be noted, however, that clines can also reflect neutral processes, particularly isolation-by-distance (Vasemägi 2006). Testing for associations between loci and environmental gradients is therefore important to substantiate assumptions of an adaptive basis of the clines, e.g. using landscape genomics and genetic association tests (Joost et al. 2007; Coop et al. 2010; Manel et al. 2010; Stranger et al. 2011). There are no generalized test frameworks available for analysing temporal shifts of clines, but a study of Drosophila subobscura by Balanya et al. (2006) serves as an illustrative example (this and related studies are reviewed in more detail later in the paper). They analysed changes of clinal patterns of chromosome polymorphisms coinciding with increasing temperatures resulting from climate change. First, they constructed a ‘temperature index’ based on PC1 of principal component analyses of monthly temperatures for each locality and time period. Next, they calculated a ‘chromosome index’ based on PC1 of principal component analyses of chromosome polymorphisms, again for each locality and time period. The ‘chromosome index’ was inversely correlated with both latitude and the ‘temperature index’, indicating an association between climate and polymorphisms. Both the ‘temperature index’ and the ‘chromosome index’ had increased significantly over the study period, documenting climate change and genetic change. Finally, shifts in the ‘chromosome index’ were parallel to shifts in ‘temperature index’ for almost all localities, suggesting genetic change in the direction of ‘warm-adapted’ polymorphisms.
Do we then have the required statistical toolbox for monitoring adaptive change at molecular markers? Much can be done with the statistical tests already at hand, but the field would clearly benefit from tests specifically aimed at genetic monitoring, i.e. temporal analysis. The FST-based tests listed in Table 1 use simulations to generate expected confidence intervals of differentiation for neutral loci. These simulations are based on demographic models in which geographically different populations exchange migrants, whereas genetic monitoring involves temporal sampling of the same populations. It would therefore be preferable to develop outlier tests that specifically simulate temporal sampling, e.g. by incorporating estimates of effective population size, to identify loci that show more temporal divergence (and thereby lower effective population size) than would be expected by drift alone.
Monitoring phenotypic traits
Analysing change in phenotypic traits holds the clear advantage of focusing on characters of direct relevance to the specific environmental change, e.g. flowering time in plants experiencing climate change (Franks et al. 2007) or age at reproduction in overharvested fish populations (Olsen et al. 2004). Qualitative traits with simple Mendelian inheritance would be ideal for monitoring, but except for the classic case of industrial melanism in the peppered moth (Kettlewell 1961; van’t Hof et al. 2011), it is probably difficult to find such traits that are at the same time relevant in the context of a specific environmental change. The vast majority of traits will therefore be quantitative, which requires separating genetic from environmental components of phenotypic variance and separating evolution from phenotypic plasticity (Box 1). Where these traits cannot be studied directly, some studies use narrow-sense heritability (h2) estimates for the traits from other populations of the same species or even related species (Edeline et al. 2007; Nussle et al. 2009). Although this approach provides an indication of the genetic basis for the trait, we should recall that heritabilities often depend on the environments in which they are estimated (Hoffmann & Merila 1999). For instance, Larsson (1993) obtained h2 estimates for tarsus length in the barnacle goose (Branta leucopsis) ranging from 0.16 (and not significantly different from zero) under poor growth conditions to 0.67 under good conditions. Although there are no universal patterns of h2 increase or decrease in stressful environments, several cases have nevertheless involved a drastic decrease in h2, presumably because of higher environmental variance as stress conditions increase (Hoffmann & Merila 1999). In such instances, studies that report phenotypic change and at the same time rely on h2 estimates measured under favorable conditions could arrive at erroneous conclusions; phenotypic change may be ascribed to adaptive genetic change when it may in reality reflect phenotypic plasticity. The degree to which heritability estimates differ across environments is expected to vary among specific traits, but the recommendation is to estimate h2 for the specific populations and environments to the greatest possible extent, or at least to obtain h2 estimates for more than one population and environment.
Testing for selection at phenotypic traits
Genetic monitoring of adaptive change at the phenotypic level, just as for the molecular level, involves testing for selection on ecological time scales. We summarize some of the methods and tests in Table 2.
Table 2. Tests and approaches for genetic monitoring of adaptive change using phenotypic traits
Potential problems for genetic monitoring
Differentiation at quantitative traits can be estimated as: QST = σ2GB/ (σ2GB + 2σ2GW), where σ2GB and σ2GW represent additive genetic variance between and within samples, respectively. QST is compared to FST at molecular markers. If QST > FST, this may indicate diversifying selection. Although designed for geographically different populations, it is applicable also to temporal analysis
Estimates of additive variance for trait(s) obtained in common garden set-up. Estimates of FST obtained from molecular markers. The ideal design should consider at least two test environments resembling the environmental conditions at the time periods considered
Logistically demanding. Continuous monitoring possible by rearing individuals every few generations under identical environmental conditions to estimate additive variance. Better suited for organisms where seeds or resting eggs from different time periods can be resurrected. QST–FST is controversial, particularly due to some studies estimating phenotypic or broad-sense genetic variance instead of additive variance (Pujol et al. 2008)
If t denotes the number of generations between two points in time, z the mean phenotypic change, h2 the heritability, and σ the standard deviation of the trait value, then the maximum expected effective population size compatible with a pure drift scenario, N*, can be estimated: N* = (1.96)2h2t/(z/σ)2. If the actual effective population size, Ne > N*, then the hypothesis of neutrality is rejected (at a 5% significance level)
Requires estimates of Ne, heritability and phenotypic data from two points in time. If Ne cannot be estimated, then as a minimum it can be assessed if N* represents an unrealistically low Ne value
Pedigreeing and estimation of quantitative genetics parameters using Animal Model
By pedigreeing and measurement of quantitative traits, quantitative genetics parameters are estimated using the Animal Model, and temporal change of VA can be assessed. Using estimates of life-time reproductive success, selection gradients and differentials can be estimated (Lande & Arnold 1983), and the predicted response to selection can be calculated from the Breeder’s Equation (Falconer & McKay 1996): R = h2S, where R denotes evolutionary response measured in standard deviations and S denotes selection differential
Requires pedigree and measurements of relevant quantitative traits
Logistically highly demanding as populations must be pedigreed over several generations. Unsuitable for organisms with large population sizes, e.g. insects, or species that are continuously distributed over large geographical areas
A major challenge of most of the approaches concerns documenting the heritable basis of the traits while at the same time keeping in mind that h2 estimates may differ among populations and environments, as discussed above. Moreover, it should be kept in mind that h2 is defined as VA/VP, where VA denotes additive genetic variance and VP total phenotypic variance. If a trait is under directional selection, then VA is expected to decrease and h2 may eventually approach zero. Hence, what should really be of interest is the genetic basis of temporal phenotypic variance and temporal change of VA. One approach controls for this issue, namely studies based on pedigreeing of populations and estimation of quantitative genetics parameters using the Animal Model (Kruuk 2004; Garant & Kruuk 2005; Pemberton 2008; Wilson et al. 2010). Some of these studies reveal the importance of monitoring not only phenotypic change, but also change in VA as the environment changes.
As an example illustrating these issues, the classical Bergmann’s rule predicts that mean body size in ectotherms will increase with increasing geographical latitude because of energetic adaptations to colder climates. This rule raises the possibility that decreasing body size and shifts in ‘Bergmann’s clines’ could reflect adaptive responses to global warming (Millien et al. 2006). Teplitsky et al. (2008) tested this hypothesis in a red-billed gull (Larus novaehollandiae scopulinus) population that had been monitored for 47 years. Even though temperature had increased, body size had decreased, and h2 of body size had been estimated at 0.33 and 0.27 for males and females, respectively, VA had not changed over the time period. Hence, the decreased body size was interpreted as reflecting phenotypic plasticity rather than an adaptive genetic response.
Criteria for demonstrating adaptive genetic change
How should we judge the evidence supporting adaptive genetic change? We define six criteria needed to convincingly demonstrate adaptive evolutionary change based on analyses of either quantitative traits or molecular markers (Table 3). In particular, we should ensure that (i) suitable genetic variation exists; (ii) the monitored traits or genes are relevant to the specific environmental stress; (iii) traits or genes are analysed over time; (iv) selection is tested; (v) shifts in traits or allele frequencies coincide with the changes expected in response to the environmental change in question and (vi) we do not infer adaptive genetic change when simple replacement by a genetically different population has instead occurred.
Table 3. Criteria for demonstrating adaptive genetic responses to environmental change
Approach for demonstrating criteria
For molecular markers:
For quantitative traits:
1. Demonstrate that suitable genetic variation exists
a) Analyse candidate loci previously demonstrated to be under selection b) Identify loci (with known or unknown functions) suggested to be under direct or hitch-hiking selection on a spatial scale, e.g. using genome scan approaches (Storz 2005)
Identify traits likely to be under selection. Is there genetic variation for the trait in the original population? a) Estimate additive genetic variance in the wild (Kruuk 2004). Ideally, estimate h2 as the environment changes as part of the monitoring design b) Use ‘common garden’, reciprocal transplant, and/or field experiments to demonstrate genetic vs. environmental bases for trait variation (Lynch & Walsh 1998) c) As a last resort, use traits where heritability has previously been demonstrated in other populations of the same species
2. Link this genetic variation to a specific environmental stress
Do we expect the gene to be under selection due to the specific environmental stress? This could be based on prior knowledge, the expected function of the locus, or empirical data For example: a) Analyse candidate loci of specific relevance to the environmental stress, such as MHC genes involved in immune response (Bernatchez & Landry 2003) or Hsp genes involved in thermal stress response (Sørensen et al. 2003) b) Test for associations between genetic variants and environmental variables associated with a specific environmental stress (Manel et al. 2010)
Do we expect the trait to be under selection due to the specific environmental stress? This could be based on prior knowledge, the expected function of the trait, or empirical data For example: a) Do traits covary in the expected way across experimental or field populations that vary in this stress? b) Does this trait variation enhance fitness?
3. Test genetic change over time
Do the identified allele(s) vary over generations? a) Initiate a continuous sampling program starting at the present b) Sample retrospectively using archived samples, e.g. herbarium specimens, bones, skins, feathers, or fish scales (Wandeler et al. 2007; Leonard 2008; Nielsen & Hansen 2008)
Do the identified trait(s) vary over generations? a) Initiate a continuous sampling program starting at the present. Alternatively, pedigree analyses may serve to track variation among families over generations b) Track trait change by growing out historical and newer individuals from seeds or eggs into a common environment (resurrection studies) c) If resurrection is impossible, track historical changes in traits known to have a strong genetic basis
4. Test that selection has occurred
Do the changes in allele frequency reflect selection? a) Conduct specific test for selection, e.g. using one or more of the methods listed in Table 1 b) Test a specific hypothesis, e.g. expected change of clinal patterns at traits due to a certain environmental change
Do the changes in trait values reflect selection? a) Conduct specific analysis/test for selection, e.g. using one or more of the methods listed in Table 2 b) Test a specific hypothesis, e.g. expected change of clinal patterns at loci due to a certain environmental change
5. Link the observed genetic change(s) to selection attributable to a particular environmental factor
Do observed shifts in a trait or allele frequencies reflect selection due to the identified environmental factor? Assess or test that the observed adaptive change coincides with the specific environmental change. If possible, obtain information on many environmental parameters to rule out confounding variables
6. Conduct tests to rule out simple replacement – the possibility that a more adapted population replaced the original population
a) Demonstrate genetic continuity over time using molecular markers, e.g. by estimating temporal differentiation. If possible, analyse several populations and conduct a hierarchical FST-analysis to document that differentiation among populations exceeds differentiation among temporal samples within populations b) If immigration from other populations is unlikely, e.g. due to isolation on an island or in a closed lake, then this should be stated explicitly
Collectively, these defined criteria represent a stringent set of conditions. In the following, we present some empirical cases that illustrate different approaches to genetic monitoring of adaptive change and provide examples of how they have met the criteria listed in Table 3. We also review a larger body of studies and assess the extent to which they have met the criteria (Tables S1 and S2, Supporting information).
Response to climate change in Drosophila
Some of the most convincing results suggesting adaptive responses to ongoing climate change are derived from wild populations of D. melanogaster (Umina et al. 2005) and D. subobscura (Balanya et al. 2006). These studies represent exemplary cases of genetic monitoring of adaptive change and address almost all of our criteria (Table 3). Both species display latitudinal (i.e. North–South) clines in both phenotypic traits thought to reflect thermal adaptation and genetic markers involving well-characterized genes and chromosome inversions. These clines appear to reflect temperature-related selection as supported by laboratory experiments (Umina et al. 2005). They also provide evidence that the polymorphisms are relevant candidates for being under selection (criterion 1) and that links exist between these polymorphisms and a specific environmental factor (criterion 2). Umina et al. (2005) further demonstrated temporal genetic change (criterion 3) in D. melanogaster introduced to Australia approximately 100 ya by analysing clines at two enzyme-coding loci and two chromosome inversions observed in 1979–1982 and 2002–2004. For one enzyme locus and one inversion, no significant temporal change was evident, but at the alcohol dehydrogenase locus and the other inversion, there were significant shifts in the clines even as the slopes remained unchanged. The authors confirmed that temperature had increased over the monitoring period and moreover that the shifts occurred in the direction expected if populations were responding evolutionarily to those increased temperatures. Hence, by testing a previously stated hypothesis (criterion 4), the study provided evidence for genetic change resulting from selection. Of interest, these cline shifts corresponded to latitude shifts of 3.9° and 7.3°, respectively, considerably more than the shift expected from changes in individual climate change factors (increasing temperature, decreasing rainfall and decreasing humidity). The authors interpreted this outcome to mean that these populations are responding to a combination of climatic factors rather than to single factors acting independently. Thus, the authors linked the observed adaptive response to a specific environmental change (criterion 5). They presented no evidence, however, to confirm that these changes reflect adaptive responses within populations (as opposed to immigration of individuals from other populations; criterion 6). Nevertheless, the Australian D. melanogaster clines arose de novo since the introduction of the species ca. 100 ya, confirming that temperature-related adaptation can proceed rapidly.
In terms of both design and results, the work of Balanya et al. (2006) with D. subobscura resembles that of Umina et al. (2005). Balanya et al. took advantage of the fact that D. subobscura is native to Europe but has been introduced to North and South America, allowing them to replicate their results and demonstrate recent cline shifts in response to climate change on all three continents.
Rodriguez-Trelles & Rodriguez (2007) subsequently criticized this study, arguing that the study design could not rule out whether seasonal cycles of allele frequency change created the apparent cline shifts. Balanya et al. (2007) later presented empirical data refuting this criticism. Nevertheless, the criticism raises an important issue for monitoring adaptive responses to climate change in organisms with short generation times. Such studies should control for possible seasonal cycles of selection regimes and evaluate how climate change alters seasonal timing. Life histories could also possibly exist where individuals from panmictic populations disperse to environmentally different foraging areas where they are subject to diversifying selection. However, this within-generation response is obliterated following each event of panmictic breeding. Such a scenario may be particularly relevant in marine fishes with high dispersal potential (Nielsen et al. 2009a), and particularly in panmictic species like the European eel (Anguilla anguilla; Als et al. 2011).
Selection imposed by hunting and fishing
Human predators represent perhaps the most significant factor modifying traits in most exploited species (Allendorf & Hard 2009; Darimont et al. 2009). One approach for studying selection imposed by hunting involves pedigreeing wild populations and estimating quantitative genetic parameters using the Animal Model (Table 2). Using such a setup, Coltman et al. (2003) tracked individual phenotypes in a pedigreed bighorn sheep (Ovis canadensis) population in Alberta, Canada, under pressure from trophy hunting (focusing on the horns) for 30 years. They used microsatellite paternity analysis to confirm the high heritability of horn and body size. They also documented significant declines in horn size, an apparently adaptive evolutionary response to hunting. Body weight had also declined, representing a correlated response. Such results suggest that selective hunting can substantially alter prey characteristics over relatively short periods of time.
Rutter (1904) wrote more than a century ago that ‘A large fish is worth more on the market than a small fish; but so are large cattle worth more on the market than small cattle, yet a stock raiser would never think of selling his fine cattle and keeping only the runts to breed’, expressing his concern about selective fishing for large Sacramento River salmon. Fisheries-induced evolution has thus long been recognized. Modern intense size-selective fishing favours faster-growing, early-maturing genotypes, as has been observed in many exploited fish populations (Jørgensen et al. 2007; Sharpe & Hendry 2009). Do these changes reflect environmentally induced changes in the demography of populations or true genetic changes? A number of studies have attempted to disentangle these factors using phenotypic data [see reviews by Jørgensen et al. (2007) and Heino & Dieckmann (2008) and empirical studies by Olsen et al. (2004), Edeline et al. (2007) and Nussle et al. (2009)]. In general, these studies documented phenotypic change over time, but the genetic basis of the observed change and identification of the specific selection regime (i.e. fisheries-induced selection as opposed to other environmental change) often remains uncertain (reviewed in Table S2, Supporting information). Mesocosm experiments involving simulated selective fisheries have confirmed that this pressure can lead to evolutionary genetic change in the predicted direction and in as few as four generations (Conover & Munch 2002), but obviously this type of study involves much less complex environmental conditions than those encountered in the wild. However, we now also see attempts to estimate fisheries-induced evolution using genetic markers, moving towards the molecular level and providing tools for monitoring wild populations (Nielsen et al. 2009a).
Analyses of DNA from historical collections of scales and otoliths allow us to infer both demographic processes, such as temporal changes in migration rates and genetically effective population sizes (Nielsen & Hansen 2008), and decadal shifts at genes of adaptive value (e.g. Nielsen et al. 2009b; Hansen et al. 2010). Such candidate gene approaches can also be used to infer fisheries-induced evolution. Although growth and maturation are likely highly polygenic in fish, a few master genes may govern these processes. The ‘growth axis’ appears to be determined by a system of hormones, including growth hormone and insulin-like growth factors, as well as various promoters, inhibitors and receptors (De-Santis & Jerry 2007). The ‘maturation axis’ appears to be similarly regulated by other hormones (e.g. follicle-stimulating hormone and luteinizing hormone) and their respective receptors, promoters and inhibitors (Weltzien et al. 2004). Genetic variation in and around these genes can be identified by sequencing as a prelude to screening for polymorphisms (e.g. SNPs) in historical and contemporary population samples (Hemmer-Hansen et al. 2011).
All the caveats and criteria described previously also apply to marine organisms. Criterion 6 – ruling out replacement by a different population – is particularly important, given the potentially high vagility of marine fishes. A recent SNP-based genome scan of Atlantic cod (Gadus morhua) was used to identify genomic signatures of adaptive variation across populations and also confirmed the temporal stability of apparent selection based on analysis of historical samples (Nielsen et al. 2009b). Neutral genetic change over time was limited, confirming the genetic continuity of populations. Hence, we now have the markers, samples and conditions needed to monitor fisheries-induced evolution at the molecular level.
Resurrection ecology and contemporary evolution in Daphnia
Time and resources limit our ability to monitor phenotypic traits across generations. Plants and animals that produce diapausing seeds or eggs, however, allow tracking of genetic changes over much longer periods by ‘resurrection’ of dormant seeds and eggs (Bennington et al. 1991; Kerfoot et al. 1999; Cousyn et al. 2001; Decaestecker et al. 2007; Franks et al. 2007). By rearing individuals from historical and contemporary populations in a common environment, we can detect adaptive responses over time and estimate rates of evolution.
Water fleas such as Daphnia play crucial roles in aquatic ecosystems by grazing on phytoplankton and providing food for fish. Their diapausing eggs can be recovered and ‘resurrected’ from stratified sediments, allowing establishment of experimental populations corresponding to different time points. Cousyn et al. (2001) used this approach to study contemporary evolution over ca. 30 years in Daphnia magna in a small pond that had experienced high variation in predator abundance. The study focused on a behavioural trait, phototactic behaviour, that results in diel vertical migration: at night, zooplankton move up in the water column to feed but retreat to deeper water during the day to avoid predators. Phototactic behaviour was known to differ among D. magna populations in lakes with different fish densities, but whether contemporary evolution could occur at this trait was unknown. The prediction was that if phototactic behaviour in the presence of fish adaptively evolved within the time frame of the study, then (i) differentiation, QST, at the quantitative trait should exceed FST at neutral microsatellite loci among temporal populations resurrected from different time periods; and (ii) there should be higher plasticity in phototactic response with exposure to predators in populations adapted to higher fish densities. Plasticity was measured by comparing phototactic responses in the absence of fish predators to responses when Daphnia were exposed to a fish kairomone (a substance released by the predator that is perceived by the prey). While QST ranged from 0.19 to 0.20, FST was much smaller (<0.022), suggesting that selection had dramatically changed the phototactic response in the predicted direction. The study thereby fulfills all criteria for demonstrating temporal evolutionary adaptive change by demonstrating selection at the trait, associating the observed selection with a specific selection regime, testing an extrinsic hypothesis concerning the predicted direction of adaptive change, and demonstrating genetic continuity of the population using microsatellite markers (Table 3). This study also demonstrates that reaction norms can evolve quickly.
Resurrection ecology studies of D. magna have provided other important insights. For instance, this system has been used to analyse Red Queen dynamics by studying host–parasite coevolution across temporal resurrected populations (Decaestecker et al. 2007). Resurrected D. magna populations have also been used to assess evolution at the gene expression level. Pauwels et al. (2007) demonstrated temporally increasing levels of Hsp60 expression and estimated evolutionary rates between 0.08 and 0.12 Haldanes among different resurrected populations, probably in response to the increasing abundance of an ectoparasite.
The resurrection studies in Daphnia and other species demonstrate the power of this approach for monitoring evolutionary adaptive change. The approach is obviously restricted to organisms with diapausing seeds or eggs, but it can yield important general insights into the evolutionary potential of species and populations for adapting to altered environments. It thus remains the most feasible means for retrospective genetic monitoring of phenotypic traits, including monitoring the evolution of reaction norms and differences in gene expression.
Review of studies
We searched the literature to identify studies employing genetic monitoring to study adaptive change. This was performed by searching Web of Science for papers containing the terms genetic monitoring, adaptive change, adaptation, contemporary evolution, environmental stress, selection, time series and temporal in different combinations. Papers deemed to represent genetic monitoring of adaptive change were supplemented with other papers already known to us. In total, we identified 44 papers/studies, 15 employing molecular markers and 29 based on quantitative traits. Tables S1 and S2 (Supporting information) summarize the experimental designs and main results of the studies. We also assessed the extent to which the papers fulfilled the criteria for having demonstrated adaptive change, as defined in Table 3.
Few studies (23%) fulfilled all six suggested criteria (Fig. 1). None of the studies based on molecular markers fulfilled all criteria, but all of them fulfilled four or five of the criteria. The lowest and highest fulfillment of criteria was found among the studies based on phenotypic traits; 24% of the studies fulfilled only 1–3 criteria, but on the other end of the spectrum, another 34%, corresponding to ten studies, fulfilled all six. Of note, seven of these latter studies were based on pedigreed populations and estimation of quantitative genetics parameters using the Animal Model, underlining the usefulness of this approach.
Is there a pattern in the criteria that were not fulfilled? Approximately half of all studies (43%) and 55% of studies based on phenotypic traits failed to address criterion 6 by not ruling out simple population replacements (Fig. 2). In some studies, such as of fishes in closed lakes (Edeline et al. 2007; Nussle et al. 2009), replacement could be ruled out because of the nature of the system, although this rationale was not specifically highlighted in the papers. Other studies were based on pedigreed or very closely monitored populations, thus providing a control for immigration (e.g. Grant & Grant 2002; Coltman et al. 2003). Among the remaining studies based on phenotypic traits, only a few specifically addressed the possibility of population replacement (e.g. Cousyn et al. 2001; Decaestecker et al. 2007). In total, this pattern demonstrates that monitoring of phenotypic traits could be improved by using molecular markers to assess temporal stability within populations and/or to document that differentiation is higher among different populations than among temporal samples within populations. Nevertheless, even if the temporal stability of a population is documented, limited gene flow among populations cannot be ruled out. Even a few immigrants could introduce adaptive variation resulting in higher fitness in a changing environment. From a conservation perspective, it would be important to know if adaptive responses result from standing genetic variation within a population or from variation provided by gene flow. Unless populations are pedigreed, demonstrating genetic rescue will be difficult, but the potential for genetic rescue could be assessed by methods for estimating dispersal among populations based on molecular markers (e.g. Paetkau et al. 2004).
Thirty-four percent of the studies monitoring phenotypic traits did not test whether selection as opposed to genetic drift caused the observed changes (Fig. 2). Although such testing can be a challenge, several possible tests exist, as described in Table 2. It seems realistic that one or more of these analyses could have been applied in the cases in which selection was not tested. For instance, studies measuring rates of evolution could examine if the observed change is compatible with a scenario involving only drift (Lande 1976; see Table 2).
Only 16% of all studies reported no adaptive change (Fig. S1, Supporting information). This finding could reflect publication bias if research that identifies no adaptive response goes unpublished. We find it striking that 36% of all studies provided evidence for adaptive responses but at the same time could not link the observed responses to a specific environmental change (Fig. S1, Supporting information). These reports included studies that did not specifically test for associations between adaptive and environmental changes and those that conducted such tests but found no significant association. Because selection regimes can be complex, this pattern emphasizes the importance of detailed collection of environmental data to identify the specific selection forces involved.
In total, only a minority of the reviewed studies fulfilled all the criteria needed to convincingly demonstrate adaptive change via genetic monitoring. We therefore recommend that future studies be specifically designed to address all criteria for demonstrating adaptive change, such as those provided in Table 3. If, for some reason, not all criteria can be fulfilled, any that were not met should be explicitly highlighted in publications, and conclusions should obviously be tempered accordingly.
Genetic monitoring of adaptive change as a tool in conservation biology
The reviewed studies collectively demonstrate that monitoring adaptive genetic change in field populations is feasible and thus could be used as a practical conservation tool. In addition to general questions concerning adaptability and adaptive responses of populations, we would like to draw attention to the following potential applications of genetic monitoring.
1 Where populations are declining and show evidence of little adaptive response, genetic rescue can be considered as a conservation option. Genetic rescue has so far been used primarily to introduce new variation into populations suffering from inbreeding depression (e.g. Hedrick 1995; Westemeier et al. 1998). This application has generally proven successful, although caveats concerning outbreeding depression and swamping of indigenous adaptive variation have also been raised (Tallmon et al. 2004). In addition to providing evidence for low adaptability of specific populations, genetic monitoring could also assist in identifying potential donor populations. Thus, where several populations are monitored simultaneously, those showing adaptive responses could be considered as candidate donors for genetic rescue of other declining populations showing no adaptive responses. With the ongoing integration of genomics methodology into population and conservation genetics (Allendorf et al. 2010; Ouborg et al. 2010), our understanding of the genomic basis of adaptation to various kinds of environmental change could even advance to an ability to predict which particular source populations, or even alleles at specific loci, might most benefit a local, non-adapting population. This potential application raises the possibility of contributing not only more genotypes but also genotypes enriched in just the kinds of genetic variation most likely to benefit the population at risk.
2 Genetic monitoring could identify harvested populations suffering from increased lag load (Box 2) and thus potentially approaching collapse. This information could be used for managing exploited populations of fish and wildlife by regulating harvesting (Olsen et al. 2004). A strict demonstration of lag load is complicated but might, for example, involve genetic monitoring along with monitoring of demographic trends. Alternatively, the theoretically maximum sustainable rate of evolution of <0.1 Haldanes per generation (Lynch & Lande 1993; Burger & Lynch 1995) could serve as a threshold, and evolutionary rates approaching this magnitude should be cause for concern.
3 Which species and populations will not adapt fast enough? This is a central question when forecasting the consequences of environmental change, such as global warming. Genetic monitoring of adaptive change could provide predictive tools for this purpose. Theoretically, we would predict that organisms with features such as small populations, long generation time, asexuality and complex interactions with their environment would adapt slowly to environmental change. However, for organisms that show intermediate values with respect to these traits, predictions will be more difficult. There may also be many more as-yet-unidentified life-history and demographic features that could influence adaptability to a certain environmental stress. A comparative approach based on genetic monitoring of several species experiencing similar environmental change but representing diversity in a variety of traits could highlight the most important factors involved in adaptability and thus improve our basis for forecasting.
Our review of studies reveals that genetic monitoring of adaptive change is already a diverse field involving many different approaches. We particularly foresee important future developments in three directions, as follows.
First, most genetic monitoring studies based on quantitative traits have focused on a few traits individually. In reality, quantitative traits within an organism show various degrees of correlation that can be described with a matrix of additive genetic variances and covariances, the G-matrix (Lande 1979). Depending on the strength of correlation and selection acting on the different traits, there may be constraints on adaptation of individual traits. Hence, the outcome of selection on a single trait may not be readily predictable, an important consideration when interpreting results of genetic monitoring. Moreover, the G-matrix itself can evolve (reviewed by Steppan et al. 2002; Arnold et al. 2008), and reconstructing and monitoring the G-matrix would therefore be of considerable scientific interest. Doing so would primarily be feasible for species that are closely monitored and pedigreed or for organisms prone to resurrection designs.
Second, we have highlighted the utility of historical samples for retrospective monitoring of molecular markers at decadal scales. A logical next step would involve analysis of ancient DNA, potentially extending time scales by thousands of years. This field has evolved dramatically within the past decade, although studies of ancient DNA still involve formidable tasks in terms of avoiding contamination problems, accounting for DNA damage and estimating the age and verifying the authenticity of samples (Willerslev & Cooper 2005).
Obtaining sufficient sample sizes is also a problem, not least because of the low amounts and quality of DNA in ancient specimens. As an example, Jaenicke-Despres et al. (2003) analysed allelic variation at three functionally important nuclear loci in domesticated maize (Zea mays) from two archeological sites in Mexico and New Mexico, covering a time span from 4300 to 600 ya. Alleles characteristic of contemporary domesticated maize were already present in the oldest samples, but on the other hand, alleles characteristic of wild maize still remained in 2000-year-old samples, indicating that domestication was not yet complete at that time. Although this study could be characterized as genetic monitoring of adaptive change, the low sample size precluded statistical analysis and did not allow for detailed reconstruction of the domestication process. However, recent technical developments such as massively parallel hybridization capture that allows for retrieving low copy numbers of highly degraded DNA could increase the possibilities for obtaining larger ancient DNA sample sizes (Burbano et al. 2010).
Finally, technical advances in genomics hold enormous potential for genetic monitoring. During the past five years, population genomics studies of non-model organisms have increased from hundreds to thousands of markers, predominantly SNPs (Namroud et al. 2008; Bradbury et al. 2010; Hohenlohe et al. 2010). With the advent of reduced-representation genome-wide sequencing (Davey et al. 2011), it is now feasible to monitor genomes rather than genes, and whole genome resequencing (Li et al. 2009) at the level of population samples is a realistic future goal. Although such analysis involves considerable bioinformatics-related effort, monitoring of genomes can provide unprecedented information about the genes involved in adaptive responses to specific environmental changes. Recently, we have even seen an example of whole-genome resequencing based on an archeological ancient DNA sample (Rasmussen et al. 2010), potentially making it possible to monitor genomes over centuries or millennia.
In conclusion, our review shows that monitoring adaptive change has already yielded important results demonstrating if and how organisms can adapt to human-altered environments. By proposing a framework of criteria to use in monitoring studies, we hope to further encourage awareness of methodology and experimental designs. With the continuing developments in quantitative genetics, bioinformatics and population genomics, we envisage a bright future for genetic monitoring of adaptive change and foresee that it will not only increase our understanding of contemporary evolution but also provide novel tools for practical conservation of biodiversity.
This work was conducted as part of the ‘Working Group on Genetic Monitoring: Development of Tools for Conservation and Management’ (PIs Fred Allendorf and Mike Schwartz) supported by the National Evolutionary Synthesis Center (NSF #EF-0423641) and the National Center for Ecological Analysis and Synthesis, a Center funded by the U.S. National Science Foundation (NSF #DEB-0553768), the University of California, Santa Barbara and the State of California. MMH was further supported by a travel grant from the European Science Foundation network ConGen: Integrating Population Genetics and Conservation and by the Danish Council for Independent Research, Natural Sciences (grant 09-072120). We thank Louis Bernatchez and four anonymous referees for numerous constructive suggestions and comments and Peter Foged Larsen, Cino Pertoldi, Torsten Nygaard Kristensen, Nina Therkildsen, Volker Loeschcke and Jesper Bechsgaard for valuable discussions.
This paper represents one of the outcomes of the NCEAS/NESCent Working Group onGenetic Monitoring (GeM) (PIs F.W. Allendorf and M.K. Schwartz). The authors share a common interest in evolutionary biology and conservation genetics. This specifically includes how to analyze populations repeatedly over time (genetic monitoring) in order to explore the temporal dimension of genetic population structure and to estimate population abundance, demographic parameters and adaptive responses.