A MarineOmics overview of methods to detect signatures of selection

Created using HackMD and tablesgenerator

Target audience

This page is designed for beginners (with some basic undergraduate-level of understanding of evolutionary biology) and provides a broad introduction to different methods commonly used to study natural selection in wild populations at a microevolutionary scale. On this page we will explore how techniques like outlier analyses, genotype-environment analysis (GEA) and genome-wide association studies (GWAS) can help us study natural selection. To learn more about specific methods, read through the suggested readings in each section and visit the tutorials where links are provided.

Introduction

This page provides an overview of different methods for detecting microevolutionary change, specifically identifying genetic loci that may be subject to selection. Natural selection is the differential survival and reproduction of individuals based on their inherited traits. Selection acts on observable characteristics (phenotypes), changing the prevalence of different gene variants (alleles) within a population over time. This leads to changes in allele frequencies over time (generations) and adaptation to spatially heterogeneous environments which may help a population better survive and reproduce in a specific environment. This is particularly relevant in the field of marine ’omics, where understanding a population’s ability to adapt to changing environments, such as climate change or habitat loss, is crucial for preventing extinction.

The methods discussed on this page focus on the microevolutionary scale and examine changes that occur within populations over a relatively short period. This is in contrast to the study of macroevolution, which examines large-scale evolutionary patterns over much longer timescales (i.e. variation in conservation or non-synonymous mutations). The methods described here are used to identify loci which may be subject to selection and are typically applicable to use for any type of population or individual genomic data (e.g. Whole Genome Sequencing, RADseq, etc). Some of the methods can be applied to datasets where you have phenotypic or fitness data (i.e. GWAS for growth, survival, phenotype, etc.), while other methods do not require any data related to either the environment or phenotypes at all!

Please note, if temporal samples are available, other approaches can be used to track allele frequency changes over time. For instance, (Buffalo and Coop (2020)) describe methods for estimating the genome-wide contribution of selection to temporal allele frequency change, providing a framework for leveraging time-series data to detect selection more directly. These temporal methods allow for a more explicit examination of how selection shapes genetic variation across generations, which is not possible with single-time point datasets (Brennan et al. (2022)).

On this page we describe three classes of methods to identify loci under selection at a microevolutionary scale using samples collected from a single time point across space/environments. .

1. Outlier Analyses identify extreme deviations from what would be expected under neutral evolution.
2. Genotype-Environment-Analysis (GEA) links genomic variation to environmental gradients.
3. Genome-Wide Association Studies (GWAS) examine genetic variants and identify those statistically associated with a particular trait. When a specific phenotype shows a significant association with a particular loci, gene or region, it suggests that this genetic variation may be under selection.

Which method should I use to find loci under selection?

The method you choose will partly depend on the data that you have available: Outlier analyses require only genetic data, but GEA requires environmental data across sampled populations, and GWAS requires trait data for each genotyped individual. Addtionally, the biology of your system is always important to consider before conducting specific analyses, and no matter which method(s) you use, it is recommended that comparisons of outlier vs non-outliers (neutral loci) are performed (Riginos et al. (2016)) as a kind-of “validation” step, where we would expect loci under selection (i.e. outliers) to demonstrate a stronger signal of environmental dependence than non-outliers. However, a common but conservative approach taken by researchers performing outlier detection is to apply multiple test methods and identify outlier SNPs that are overlapping across them. For example, some studies perform a differentiation outlier test first (see Outlier Analyses Section) and then use only the resulting outliers as input for the Genome-Environment Associations (Rellstab et al. (2015)).

Importantly, to conduct outlier analyses you do not necessarily need to have neutral genetic differentiation (genetic variation among populations primarily driven by random processes like genetic drift and migration) in your data. For example, marine invertebrates with pelagic larval dispersal and high gene flow which show very low to no detectable population structure at neutral alleles across their range (i.e using STRUCTURE, ADMIXTURE, PCA) still show signals of adaptation. Other examples include Pacific purple sea urchins (((petak2023whole-cp?));(Rumberger et al. (2023))) or scallops (Van Wyngaarden et al. (2017)).

Outlier Analyses

What are Outlier Analyses?

Genetic differences among groups may be driven by a combination of genetic drift, demography and/ or selection. Each of these processes can directly change the allele frequency of a population. However, alleles that are under selection and influence fitness are expected to occur at frequencies different to those affected only by neutral processes. For example, loci under divergent selection (and those physically linked to them) exhibit stronger differentiation than neutral regions and so we might expect these loci to be more differentiated between populations/groups. We can identify these differences in allele frequencies among different populations/environments using outlier tests. Outlier tests for selection assume a null model of neutrality, where allele frequencies drift randomly without selection pressures. Loci with significantly higher or lower differentiation (i.e. Fst) between populations/groups than expected under a neutral model are flagged as potential targets of selection. Specifically, while selectively neutral loci are expected to form a normal frequency distribution, when alleles increase fitness (positive selection) they can be expected to be observed at higher frequencies and where they decrease fitness (negative selection) they can be expected to be observed at at lower frequencies (Hoban et al. (2021)). These are “outliers” or “putatively adaptive loci” which may be candidates of selection.

There are two commonly applied approaches to outlier analysis; Fst-based analysis (Figure 1) and PCA-based analysis (Figure 2). Both are used to identify loci under selection, but they differ in their underlying methods. Fst is a standardized measure of the variance of allele frequencies among populations (Wright (1949)). By directly comparing genetic differentiation between populations/groups to neutral expectations, Fst outlier tests can identify loci which may be under selection (Figure 1). In contrast, PCA-based methods use Principal Component Analysis (PCA) to capture population structure and can identify loci with extreme loadings on principal components as potential outliers (Figure 2).

Figure 1: Example of an Fst-based analysis using OutFLANK. In this dataset, 1000 individuals were collected from across 39 populations spanning an environmental gradient. Each dot is a SNP and its location in the genome plotted on the X-axis. The Y-axis is the Fst for each SNP. Higher Fst values are indicative of larger genetic differentiation and candidate SNPs under selection are highlighted in pink. Source

Figure 2: Example of a PCA-based analysis using pcadapt. In this dataset, a total of 150 individuals from three different populations were genotyped at 1,500 markers. Population structure was identified for three discrete groups of individuals, matching population assignment expectations (A). SNPs that are significantly associated with PC1 and PC2 are given higher weightings and are used as input for a Manhattan plot, where the SNP location in the genome is plotted on the X-axis and the weightings (or -log pvalue) are plotted on the Y-axis. SNPs that have higher weightings will have higher -log pvalues and are candidate SNPs under selection among populations (B). Source

General Method Overview:

Step 1 - Filtering your dataset

Non-selective evolutionary forces can affect the distribution of outlier loci from loci across a genome (Lotterhos and Whitlock (2014)), therefore, the first step is to identify the underlying distribution that is expected under neutral processes by answering the question; how much variation in among loci would be expected (i.e. the null distribution of Fst) in the absence of selection? One must first create a null distribution using a set of SNPs (single nucleotide polymorphisms) that accurately model the distribution expected under neutrality. An accurate null distribution typically requires some filtering of a set of SNPs, and in many cases, the best set of SNPs for neutral parameterization is not the entire set of SNPs that you want to evaluate for outliers. Common filtering steps could include a minor allele frequency/count, removing missing data and related individuals, genotype depth and quality, and removing SNPs in linkage disequilibrium (LD). As a general guide, we recommend the following SNP filtering tutorial. Deciding on how much filtering is required can be tricky and can feel subjective. We recommend visualizing the filtering steps using a PCA, which will elucidate many common biases in the dataset (see the MarineOmics page on PCAs). We recommend using the same set of SNPs that you use for estimating population structure for neutral parameterization (which is often a set of SNPs not in LD). Once neutral parameterization is performed on one set of SNPs, for most programs that model can be applied to the larger filtered SNP dataset.

Step 2 - choose an outlier method based on the biology of your system!

When choosing an outlier method, it is important to understand what biological assumptions are being ignored/violated and whether the model accounts for the demographic history when estimating divergence among populations (see Table 1). For example, early models such as fdist and fdist2 (Beaumont and Nichols (1996)) assume populations are all equal in size and there is equal probability of migration and gene flow among all populations (even if some populations are located further away than others). This method produced a lot of false positive outliers because more realistic scenarios violated these simple assumptions (Lotterhos and Whitlock (2014)). An improvement over this approach was implemented in the program BayeScan where each population has its own neutral model (Foll and Gaggiotti (2008)), but like fdist/fdist2, BayeScan also suffers from a large false positive rate due to its violated assumptions of non-independence among populations. Newer outlier methods attempt to account for population demographic history and independence in different ways. For example, pcadapt uses PC loadings (insead of Fst distributions) which are meant to be less reliant on assumptions about the underlying demography (Duforet-Frebourg et al. (2015)). However, PC loadings can be influenced by recombination variation, so it is important to use a set of SNPs thinned for linkage disequilibrium for neutral parameterization (Lotterhos (2019)). These estimates of population structure are included as covariates in the final Outlier Analysis model. See Table 1 for more information on how each method accounts for population demographic history.

Population structure is also important to consider. For example, if you discover only one genetic cluster (K=1 in STRUCTURE or Admixture plot), perhaps any of the methods in Table 1 would work well for your data. On the other hand, if there is a lot of population structure (i.e., K>1), then a method that accounts for demographic history should be chosen. Additionally, consider what assumptions are made in the tutorials that you use as a resource. Does the tutorial you are using assume that your data is normally distributed? Multiple tutorials have been added below to help with comparing between different assumptions made for different kinds of data.

Many researchers will identify outliers using two or three of these methods and look for overlap between them (Silliman (2019)). This is a conservative approach, and researchers should keep in mind that they may not find much overlap among different analyses because of the different assumptions made by each method.

Further Recommended Reading:

• (Whitlock and Lotterhos (2015)) (Outflank)–method comparison
• (Silliman (2019)) –empirical study with method comparisons
• (Lotterhos and Whitlock (2014))(Fst outlier methods)–method comparison
• (Lotterhos (2019))(pcadapt, Outflank, Baypass)–method comparison
• (Lotterhos, Fitzpatrick, and Blackmon (2022))–method validation
• (Capblancq et al. (2018)) (GEA & pcadapt)–method comparison
• (Luu, Bazin, and Blum (2017)) (pcadapt)–method details
• (Hoban et al. (2016))–helpful review and suggested best practices for statistical analyses
• (Robertson (1975)) – theory
• (Ahrens et al. (2018)) –review, summary of findings from outlier scans, and suggested best practices for experimental design

TABLE 1. Common methods for identifying outliers

Common methods for identifying outliers and the associated approach for calculating divergence, the biological assumptions, and how demographic history is accounted for in the model. We also provide a link to public tutorials for each method.

Model & Approach	Demographic History	Assumptions / Known issues	False Positive Rate	Tutorial	Citation
fdist & fdist2 (Fst; using Arlequin software)	* accounts for loci with different heterozygosities and with finite sample sizes per subpopulation	* assumes populations are equal in size and migration rates * non-independence among populations	High (Lotterhos and Whitlock (2014)) (Narum and Hess (2011)) (Pérez-Figueroa et al. (2010)), (Whitlock and Lotterhos (2015))	Outdated due to high false positive rates	(Beaumont and Nichols (1996))
Bayescan (Fst)	* populations defined ahead of time * Dirichlet-multinomial distribution: each population has its own parameter for neutral model	* does not allow for the non-random correlations among pairs of populations that cause the non-independence of sampling * slow run time (Luu, Bazin, and Blum (2017))	Low to high depending on the demography	Benestan tutorial (2020)	(Foll and Gaggiotti (2008))
PCAdapt (PC loadings)	* Latent factors to estimate values of K * uses PC loadings (less reliant on assumptions about the underlying demography) (Duforet-Frebourg et al. (2015)) * Important to use a set of SNPs thinned for linkage disequilibrium for neutral parameterization (Lotterhos (2019))	* individuals not sorted into populations ahead of time * Performs well in populations with high admixture	Low to medium (Capblancq et al. (2018)), (Luu, Bazin, and Blum (2017)) Might not be suitable for populations with complex demographic histories.	Luu, Privé and Blum tutorial, UC Davis Marine Genomics course tutorial	(Duforet-Frebourg et al. (2015)) (Silliman (2019))
Outflank (Weir and Cockerham’s Fst)	* populations defined ahead of time	* Estimates the number of K genetic clusters to inform a null model of Fst distributions * Performs poorly in populations with admixture and PoolSeq samples * Affected by finite sample sizes	Low, but suffers from low power	Lotterhos Tutorial (2018), UC Davis Marine Genomics course tutorial	(Lotterhos and Whitlock (2014)), (Schlötterer et al. (2014))
FLK TF-LK	Sensitive to complex patterns of migration and sharp differences in the migration rate among populations (Bonhomme et al. (2010))	* The program FLK uses a population tree (without migration) to describe the covariance among populations and build a null model of Fst distributions.	Low (Bonhomme et al. (2010))	UoF Bioinformatics Hub Tutorial	See supplementary file 1 from (Bonhomme et al. (2010))
BayPass XTX statistics	Assumes HWE for PoolSeq data (Gautier (2015))	* Uses a covariance matrix of allele frequencies among populations (Pickrell and Pritchard (2012)) * Can be used with or without environmental data	Low	Barley-Agroclimatic-Association Tutorial, Erica Nielson Tutorial	(Gautier (2015))

Genome Environment Association (GEA)

What is a Genome-Environment Association?

The distribution of genetic diversity within natural populations is significantly influenced by environmental conditions. Along environmental gradients, many marine species demonstrate genetic variation for adaptive traits (Pinsky (2021)). Genotype environment association (GEA) analysis is a statistical framework that allows us to investigate these genotype-environment relationships and identify specific genetic markers (loci) associated with environmental variation across populations (Rellstab et al. (2015)). In a heterogeneous environment, environmental factors may act as selective pressure for local adaptation. In this case, allele frequencies at loci underlying adaptive phenotypic traits should be associated with environmental factors that act as proxies for the unobserved selective pressures (Frichot and François (2015)). GEAs are distinct from outlier detection methods previously described as they do not rely on an allele frequency distribution, but instead use the correlation between genotypes and environment for identifying both loci under selection and key climatic factors driving local adaptation. GEA analyses usually control for isolation by distance in addition to demography (i.e. ensuring that associations between genotype and environment are not driven by physical distance between environments).

General Method Overview: GEA Methods

GEA broadly refers to any methodology that assesses the association between allele frequencies and environmental variables. Therefore, you will need four things before getting into GEAs - a good sampling regime, genetic data, geo-referenced environmental variables (see Section 4. Environmental data). GEA methods commonly cited in the literature include: Redundancy Analysis (RDA), Gradient Forest (GF), Bayesian methods (i.e. Bayenv), and Latent Factor Mixed Models (LFMM).

Redundancy Analysis (RDA) is a multivariate statistical technique used to explain the variation in a response variable (e.g., genotype) by a set of explanatory variables (e.g., environmental factors). RDA is an extension of Principal Component Analysis (PCA) that relates two data matrices. It identifies linear relationships between the response variable and explanatory variables (Figure 3).

Figure 3: Example of a RDA plot. The blue vectors are the environmental predictors. Individuals are color coded by location and plotted in the ordination space based on their relationship with the axes (which are linear combinations of the predictor variables). SNPs are plotted gray in the center of the plot. Note that individual genotypes from wet and temperate British Columbia (green) are positively related to high annual precipitation (AP) and low temperature seasonality (sdT). Based on RDA results, SNPs are strongly correlated with these two precipitation values and these candidate SNPs are plotted along the AP vector out from the center of the plot. Source

Gradient forest (GF) (Ellis, Smith, and Pitcher (2012)) is a machine learning method that is a multidimensional extension of the random forest algorithm (Breiman (2001)). GF uses cross-validation to identify loci under selection, compute the importance of predictors and can identify important thresholds or change points in genetic variation along the environmental variable (Figure 4). GF can also handle non-linear relationships and correlated predictors.

Figure 4. In a study by (Nielsen et al. (2021)), GF modeling was used to infer genomic variation–environment relationships. Models were fitted to the allele frequencies of putatively neutral and outlier SNP datasets (detected with other methods) for three rocky shore species and environmental data from WorldClim and principal coordinates of neighbor matrices (PCNMs) were used as predictors in the model to account for spatial autocorrelation and unmeasured environmental variation. The figure shows cumulative importance curves based on the turnover functions for each environmental variable for the urchin (left) where the curve shape indicates the rate of change in allele frequencies along the environmental gradient, and the maximum height indicates the total turnover in allele frequencies. The authors then used the turnover functions from the GF models to transform the environmental variables into a common biological importance scale (i.e (Fitzpatrick and Keller (2015)) (left). We can see neutral (a, c, e) or outlier (b, d, f) loci where the coloured map cells indicate similar allele frequencies and show that the genomic turnover patterns of putative neutral and outlier loci generally followed the biogeographical breaks of the marine environment in the region.

Bayenv is a Bayesian hierarchical model that estimates the posterior probability of a locus being under selection. It models allele frequencies as random variables and estimates the probability of selection for each locus (Figure 5).

Figure 5: An example of a SNP where a lower allele frequency was found to be significantly associated with warmer minimum winter temperatures using Bayenv. This SNP was characterized by a high Bayes factors and extreme allele frequencies. (A) Allele frequencies and standardized minimum winter temperatures of the SNP, rs6001912 (B) shows the geographical distribution of SNP rs6001912 (Fitzpatrick et al. (2021))

Latent factor mixed models (LFMM) is a univariate regression model that includes unobserved variables (latent factors) that act to correct the model for confounding effects (i.e.population structure), which can improve power of the model when environment and demography are correlated (Figure 6) ((Frichot et al. (2013)); (Frichot and François (2015))).

Figure 6: In a study on red abalone population genomics by (Griffiths, Smith, and Whitehead (2025)), lfmm2 modeling was used to infer genomic variation–environment relationships with maximum sea surface temperature (SST) (A), Coastal Upwelling Transport Index (CUTI) (B), Biologically Effective Upwelling Transport Index (BEUTI) (C) and range BEUTI (D). The horizontal line on the manhattan plots represents the significance threshold after Bonferroni correction of 0.05 (red). Very few SNPs passed the significance threshold and they are spread across multiple scaffolds, suggesting there is weak evidence for local adaptation to these environmental gradients.

So you want to do a GEA?

So you want to perform GEA? You will need four things before getting started…

1. A good sampling regime
2. Genetic data
3. An understanding of population structure/ and/or isolation by distance (IBD) in your system
4. Geo-referenced environmental variables and a GEA method (or two)!

Experimental Design Conditions for GEA

1. A good sampling regime

Sampling design has implications for minimizing false-discoveries and for detecting true positives (statistical power) in relation to environmental selection. Compared with random or opportunistic sampling approaches, informed sampling designs based on the characteristics of the study area and species life history tend to improve the power of GEAs (Whitlock and Lotterhos (2015)) . The general consensus suggests “stratified sampling” (Dauphin et al. (2023)) - a sampling regime that is informed by a priori knowledge of neutral genetic differentiation (or geographic variation) and/or environmental representativeness to capture genetic diversity and population structure. To conduct a stratified sampling method you would collect an equal number of samples from each group, where each group is identified from environmental and population structure information ((Dauphin et al. (2023)), Figure 7). If you are unsure of the genetic structure of your species, it is recommended that you conduct a PCA or build a map of the environmental variables across your potential study area, with the aim to visualize the environmental variation across geographical sites before sampling (see section 4 on Environmental data). In this case, sampling locations that are environmentally different, but replicated within environments can help distinguish selection from neutral population structure. A key problem in this field is when unobserved variables correlate with both the primary environmental variables and the genomic data. For example, GEAs tend to have low power and high rates of false positives when environmental gradients are correlated with neutral population structure ((Whitlock and Lotterhos (2015)); (Caye et al. (2019))). Several methods have been specifically developed to account for this issue (see Table 2).

Figure 7 (adapted from Figure 1 from (Dauphin et al. (2023))): The green boundary represents the species regional range limit and the colors reflect topography. There are three main sampling design strategies that can be used in landscape genetic studies: (A) a sampling design that doesn’t rely on environmental or genetic data, (B) a sampling design that considers environmental data only, and (C) a sampling design that takes into account genetic and environmental information (Circles in box C represent sampling locations, with shadings referring to a priori known neutral genetic differences).

2. Genetic data

Any type of genetic data can be used for GEA. Whole genome data (WGS) is typical in GEA studies, though reduced representation data (i.e. RADSeq, (Xuereb et al. (2018)) is also commonly used. It is important that the genetic dataset is filtered for potential biases, such as applying minor allele frequency/count cut-offs, removing missing data and related individuals, and minimum genotype depth and quality. The same set of filtering guidelines we recommended above for Outlier Analyses, also applies here to GEA (please see Step 1 in the Outlier Analyses Section more details and recommended tutorials).

Some examples of different genomic data applying GEA methods available in the literature include:

• (Xuereb et al. (2018))–RadSeq
• (Griffiths, Smith, and Whitehead (2025)); (Rumberger et al. (2023)); (Wooldridge et al. (2024))– WGS
• (Forester et al. (2025))–Exome Capture

3. Understanding population structure in your data

Just like the outlier methods previously described, understanding Isolation by Distance (IBD) and/or population structure is critical before performing GEA analysis. IBD refers to segments of the genome shared by individuals due to common ancestry, and the degree of relatedness which can vary with distance whereby genetic similarities might be influenced by shared ancestry rather than true genotype-environment associations. Population structure, which reflects how genetic variation is distributed across different groups, can also confound results by creating artificial associations between genotypes and phenotypes. Correcting for IBD and population structure ensures that the identified associations in GEA studies are reflective of actual biological relationships, improving the reliability and validity of the findings.

Some common approaches for controlling for IBD or population structure in GEA include:

Principal Component Analysis (PCA): PCA is often used to detect and account for population structure. It reduces the dimensionality of genetic data and captures the main sources of variation. The first few principal components (PCs) usually correspond to major population substructures. By including these PCs as covariates in statistical models (such as GEA or GWAS), researchers can control for population structure and avoid spurious associations due to genetic differentiation between subpopulations.

Ancestry: Ancestry scores (like outputs from NGSadmix, STRUCTURE, ADMIXTURE) can quantify the proportion of an individual’s ancestry from different populations. These scores can be used to adjust for population structure in association analyses by including them as covariates, helping to differentiate genetic associations that arise from population demography vs actual genotype-environment relationships.

Geographical Coordinates (Latitude/Longitude): In some studies, particularly those involving environmental associations, researchers may use the geographical location of individuals (latitude/longitude) as a covariate. This helps to account for geographic gradients in environmental variables that may correlate with genetic variation.

Distance-Based Moran’s Eigenvector Maps (dbMEM): Distance-Based Moran’s Eigenvector Maps (dbMEMS) are derived from distance matrices, which reflect genetic or geographical distances between individuals. dbMEMs can be used to model spatial structure by decomposing distance matrices into eigenvectors that represent the underlying spatial patterns in genetic data. These eigenvectors are then incorporated into the model to control for spatial autocorrelation.

4. Environmental data

Lastly, you will need geo-referenced environmental data (see Table 3A and Table 3B). There are two key types of environmental data that are used; in situ or spatially interpolated data:

1. In situ data, are on-site measurements of environmental conditions usually collected directly from the field or from remote sensing devices.
2. Spatially interpolated data (e.g. oceanographic models) using different models and processes to remove missing data and errors, extend the evaluation of environmental conditions to periods and locations not covered by in-situ measurements.

Spatially interpolated data are commonly derived from the assimilation of atmospheric and oceanographic data from models, satellites and in situ observations. These ocean models usually provide accurate estimates of the state of the system, and are commonly used for GEAs because they have broad spatial and temporal resolution, and are usually publicly available and accessible. Some of the useful databases hosting past, present and future climate information can be found in Table 3A, Table 3B, and Table 3C.

The most important consideration when selecting a database is the resolution and grain of the environmental dataset. For example, due to the small spatial scale of some key regionally specific processes (e.g. coastal processes), you should aim to identify and use an ocean model with a sufficiently high spatial and temporal resolution to resolve those processes (e.g. upwelling) which may be important for selection/adaptation in your system. In this case, you might choose to use a regional model (downscaled model) which uses data from a global climate model as input into a fine-scale, regional atmospheric model with detailed bathymetry, coastal boundaries, etc.

Resources for Environmental Data:

There are plenty of existing resources for processing this environmental data

• Data extraction from the NetCDF databases

• How to extract and plot data from Worldclim

• Data extraction from the Bio-ORACLE databse

• Another data extraction method from the Bio-ORACLE database

Further recommended reading:

• (Rellstab et al. (2015)) –method comparison
• (Lotterhos and Whitlock (2015))–method comparison
• (François et al. (2016))–method comparison for false discovery rates
• (Forester et al. (2016)) (GEA)– method comparison
• (Dauphin et al. (2023))–discussion of how different types/scales of of environmental data can be used to answer different questions
• (Capblancq et al. (2018)) – method comparison (GEA & pcadapt)
• (Alvarado et al. (2022)) - example paper that uses gradientForest
• (Riginos et al. (2016)) – spatial scale in landscape genomics
• (Capblancq and Forester (2021)) –RDA
• (Selmoni et al. (2020))– example landscape genomics study
• (Assis et al. (2018))– paper on bioclimatic modelling

TABLE 2: Common methods for identifying candidate SNPs associated with environmental data (GEA)

Common methods for identifying candidate SNPs associated with environmental data (GEA), the associated approach for calculating divergence, the biological assumptions, and how demographic history is accounted for in the model. We also provide a link to public tutorials for each method.

Model & Package	Approach / Statistics	Demographic History	False Positive Rate	Tutorial / Citation
Bayenv / Bayenv2 (Command Line)	XtX statistics; Univariate correlation between environment and allele frequencies	Covariance matrix from genome-wide data controls for demography	Inflated; use PCs or Bonferroni correction	Maier 2017 Tutorial (Coop et al. (2010))
BayPass (Command Line)	XtX statistics	Pairwise covariance matrix controls for demography	Similar to Bayenv; adjust for multiple testing	XtX only, With env data, (Gautier (2015))
BayeScEnv (Command Line)	Bayesian Fst; like BayeScan, populations defined a priori	Dirichlet-multinomial model with pop-specific parameters	Not well documented	GitHub Tutorial, BayeScan PDF, (De Villemereuil and Gaggiotti (2015))
RDA (R, Vegan Package)	Multiple regression + PCA; Mahalanobis distance (Q-values)	Partial RDA includes demographic covariates	Generally low; may be sensitive to multicollinearity	KE Lotterhos Tutorial, Partial RDA, RDA GEA, (Capblancq and Forester (2021))
LFMM / LFMM2 (R, LEA Package)	Univariate LFMM (MCMC or least-squares)	Latent factors (K) estimate neutral structure	Low, but higher in complex population structure; use FDR or Bonferroni	Bay Lab Tutorial (Frichot et al. (2013))
Gradient Forest (GF)	Random forest; multivariate machine learning	Requires prior filtering (e.g., Bayenv) or spatial predictors (MEMs)	Thought to be lower; avoids multiple testing directly	GF Tutorial (Ellis, Smith, and Pitcher (2012))
Generalized Dissimilarity Modeling (GDM)	Distance-based multivariate; nonlinear relationships	Can include population structure if provided	Fewer assumptions; less affected by marker independence	(Ferrier et al. (2007)) (Fitzpatrick and Keller (2015)
Weighted-Z Analysis (WZA)	Window-based GEA; aggregated signal across SNPs	Does not explicitly model structure; looks at patterns across windows	Reported lower than BayPass and LFMM	WZA GitHub Tutorial, (Booker et al. (2024))

TABLE 3A. Oceanographic databases

A summary (not an exhaustive list) of freely available oceanographic databases to find environmental data for your GEAs

Database	Description	Link/ Reference
WorldClim	A set of global climate layers (gridded climate data in GeoTiff format) that can be used for mapping and spatial modeling.	WorldClim Website
World Ocean Database (WOD)	Includes temperature, salinity, sea surface topography, chemistry/biology optics	WOD Website
NASA’s Earth Observing System Data and Information System (EOSDIS)	Includes temperature, salinity, sea surface topography, chemistry/biology optics	EOSDIS Website
Australian Ocean Data Network (AODN)	Online network of marine and climate data resources, includes a wide range of parameters in different ocean environments collected from ocean-going ships, autonomous vehicles, moorings and other platforms.	AODN Website
NOAA Data Buoy Center (NDBC)	National Oceanic and Atmospheric Administration’s (NOAA) National Weather Service (NWS) which operates, and maintains a network of data collecting buoys and coastal stations. Can be accessed using R package rnoaa	NDBC Website
ICES Data Centre - Ocean (North Atlantic)	Oceanographic data which includes includes temperature, salinity, oxygen, chlorophyll a, and nutrients	ICES Website
Baltic Environmental Database (BED) (Baltic Sea)	Bathymetry, hydrography, nutrients, sediment data, atmospheric nutrient input, and runoff data for the Baltic Sea.	BED Website

TABLE 3B: Environmental data reanalyses

Climate reanalyses combine past observations with models and so they tend to have wide coverage and can provide data for variables and locations where we don’t have observations (i.e here)

Database	Description	Link/ Reference
Copernicus Marine	A database hosting various re-analysis products. Includes For example Global Ocean Physics Reanalysis (GLORYS12V1 (Lellouche et al. (2018))) includes GLOBAL_MULTIYEAR_PHY_001_030 which specifically provides an eddy resolving (1/12°) global ocean simulation, covering the recent period, at 1/12 degree (approx. 8km) and on 50 standard levels, in four dimensions. ERA5 provides a snapshot of the atmosphere, land surface and ocean waves for each hour from 1940 onwards (i.e reanalysis ERA5) Many code examples for accessing data using your fave language (R, python, Julia) here.	Copernicus Link (Jean-Michel et al. (2021))
Bluelink ReANalysis (BRAN) (Australsia region)	A series of ocean reanalyses; the most recent being BRAN2020 (Chamberlain et al. (2021)) BRAN is closely related to OceanMAPS, an operational ocean reanalysis system used by the Australian Bureau of Meteorology (Brassington et al. (2007))	(Chamberlain et al. (2021))
European Centre for Medium-Range Weather Forecasts	ERA5 provides a snapshot of the atmosphere, land surface and ocean waves for each hour from 1940 onwards (i.e reanalysis ERA5)	ECMWF Link

TABLE 3C: Regional projections/ Future Climate Scenarios

Examples of some present and future climate change predictions (regional, typically modeled with reanalysis products).

Database	Description	Link/ Reference
Bedford Institute of Oceanography North Atlantic Model (BNAM)	An eddy-resolving, Nucleus for European Modelling of the Ocean (NEMO) based ice–ocean coupled model. Includes past, present day and future projections.	(Wang et al. (2021)), Contact the authors for the data, .nc files.
The Gulf of Maine (GoM) Projections	Includes two high resolution numerical ocean models used to downscale global climate projections to produce four estimates of ocean physical properties in the GoM in 2050 for the “business as usual” carbon emission scenario.	(Brickman et al. (2021)) model outputs are available upon request as .nc files.
Northeastern Pacific Canadian Ocean Ecosystem Model (NEP36-CanOE) Climate Projections	Three simulations from the Northeastern Pacific Canadian Ocean Ecosystem Model (NEP36-CanOE), a configuration of the Nucleus for European Modelling of the Ocean (NEMO) V3.6. Includes a historical simulation of the 1986-2005 mean climate and two future simulations project the 2046-2065 mean climate for representative concentration pathways (RCP) 4.5 (moderate mitigation scenario) and 8.5 (no mitigation scenario).	Contact authors

Genome-Wide Association Study (GWAS)

What is a Genome-Wide Association Study?

A GWAS is used to associate some phenotypic trait with putatively causative loci/SNPs. The methods used are similar to GEA, but instead of environmental data as the explanatory variable, phenotypic traits are used. Importantly, the phenotypic and genotypic data must be tied to each individual in the study. This has primarily been used in humans to identify genetic variants associated with disease (Uffelmann et al. (2021)), but it is becoming more prominent for use in non-model organisms for how a particular trait related to climate adaptation (e.g., physiological, morphological, and life history) is under selection. For example, identifying loci associated with coral bleaching (Figure 8) (Fuller et al. (2020)) and beak size/morphology in Galapagos finches (Enbody et al. (2023)). Effect sizes of loci can also be determined based on how strongly associated variants are with phenotypic variation.

In contrast to GEA, GWAS often requires a much larger sample size (and even more so for polygenic traits). This is why many GWAS studies are now emerging that employ a low-coverage sequencing approach combined with genotype imputation ((Lou et al. (2021)); (Jørsboe and Albrechtsen (2022))),where a larger number of individuals are sequenced, increasing the power to detect significant associations. Additionally, the inclusion of a larger sampling strategy will encompass a wide range of genomic diversity to identify all of the SNPs associated with the trait, such as rare SNPs.

Similar to outlier and GEA studies, GWAS are also subject to genotyping bias issues and false positives resulting from demographic history (Tam et al. (2019)). We recommend the researcher perform some combination of filtering on their dataset, which could include removing rare SNPs, SNPs not in Hardy-Weinberg equilibrium, and SNPs that are missing from a fraction of the individuals genotypes. Go back to our “Outlier Analyses” section for more details and recommended tutorials for filtering.

Figure 8: Example of GWAS implemented using a linear regression in the program PLINK. 237 coral samples were collected across 12 reefs from the Great Barrier reef. Low-coverage WGS was performed and bleaching susceptibility was measured in each individual. Each dot is a SNP and it is plotted by its location in the genome on the X-axis. The Y-axis is the -log p value for the linear regression of each SNP and its association with bleaching susceptibility. Larger values demonstrate stronger associations. The authors identified two genes with significant loci that were associated with bleaching susceptibility. Source: (Fuller et al. (2020))

Different GWAS Sampling Strategies Will Require Different Statistical Model Designs

GWAS studies could take two approaches where traits and genotype information are collected from wild, unrelated, individuals (a population-based GWAS), or from related individuals (a family-based GWAS). In addition, these two types of GWAS could include a GxE interaction where traits are measured under different experimental treatment exposures and genotype information is collected at the end of the study. While both approaches employ similar statistical analyses, such as a linear or logistic regression depending on whether the trait is continuous (e.g., body size), or binary (alive/dead) (See Table 4), the model matrix design may require a different set up. For example, if sampling spans across a large and variable landscape, significant genotype-phenotype associations may actually be false positives due to population structure (Tam et al. (2019)). Demographic history should be accounted for as a covariate in the model. Different methods outlined in Table 4 will estimate population structure differently and the reader may want to choose a method that is best suited to the biology of their system.

Accounting for sample relatedness and ancestry is incredibly important for family-based GWAS approaches and may otherwise result in over-estimates of SNP-based heritability for the trait. In outlier and GEA, you may want to remove related individuals from the analysis, but for GWAS approaches, related individuals may increase your power to detect significant and heritable SNPs! Some GWAS designs specifically use a breeding or pedigree based approach for this reason, but related individuals must be identified first and pedigree information provided as a covariate in the model design to GWAS.

Other model matrix designs may include the addition of a fixed effect based on treatment exposures, specficially if it’s a part of the experimental design. There may be other important covariates to include in the model as random effect terms, such as the age or sex of individuals depending on the biology of the system, or independent replicant treatment exposures depending on the experimental design.

General Method Overview:

There are a variety of methods for GWAS, many that vary based on the type of sequencing or experimental design. For example, ANGSD-asso and SNPTEST are designed to work well with genotype probabilities from low-coverage data or data that has been imputed first. PLINK/PLINK2 is a widely used tool with lots of documentation. Other methods differ on the types of statistical associations employed. For example, the programs GEMMA and fastGWA were designed to efficiently (and quickly) test for associations for large numbers of individuals using linear mixed models. Methods also differ widely for how they estimate population structure and genetic relatedness among individuals. Some methods use a PCA based approach for estimating population structure (GEMMA, fastGWA, and SNPTEST). In a slightly similar approach, PLINK/PLINK2 uses classical Multidimensional scaling (MDS) and LEA uses latent factors to estimate values of K. Other programs, such as ANGSD-asso, require external programs for estimating population structure, but accept multiple inputs from PCAngsd, ADMIXTURE, or NGSadmix. For genetic relatedness, some approaches, EMMAX and GEMMA generate a kinship matrix or genetic matrix that is used as a covariate in the GWAS model. Tutorials for ANGSD-asso, PLINK/PLINK2, LEA, and SNPTEST don’t specifically touch on whether genetic relatedness is accounted for, thus these methods are not recommended for family-based GWAS approaches. Even if researchers were to use this method for population-based GWAS, we recommend researchers use external programs to estimate genetic relatedness (i.e., COLONY, CERVUS, AlpaAssign, King’s relatedness matrix) to identify related individuals to remove them from the dataset before proceeding with GWAS. For more details, see Table 4 and our further recommended reading below.

Further recommended reading:

• (Uffelmann et al. (2021))–overview of different programs available
• (Fuller et al. (2020))– GWAS example study
• (Enbody et al. (2023))–GWAS example study
• (Tam et al. (2019))–benefits and limitations of GWAS
• (Yanez et al. (2023))–Applied uses of GWAS
• (Kijas et al. (2018))–example study

TABLE 4: Common methods for identifying candidate SNPs associated with phenotypic data (GWAS)

Common methods for identifying candidate SNPs associated with phenotypic data (GWAS), the associated approach for calculating divergence, the biological assumptions, and how demographic history is accounted for in the model. We also provide a link to public tutorials for each method.

Program	Approach/ Method	Demographic History	Tutorial	Citation
BEAGLE + ANGSD-asso	Impute missing genotypes with Beagle, and use beagle output as input for ANGSD -latent model (using GLM framework) that can account for genotype probabilities for low depth seq data	* Pop. structure detected with NGSadmix or ADMIXTURE or PCAngsd programs -changes allele freq prior if pop structure is present * does not account for closely related individuals–our suggestion would be to remove these individuals from the dataset	Bay Lab Tutorial ANGSD Tutorial	(Jørsboe and Albrechtsen (2022))
GEMMA (command line)	Univariate and multivariate Linear mixed models (LMMs)	* user accounts for pop structure independently by calculating eigenvalues from user-generated PCA as covariate option * relatedness matrices calculated using either the centered or standardized genotype methods	U of Chicago Workshop Tutorial Northwestern Tutorial	(Zhou and Stephens (2012)) (Zhou and Stephens (2014))
PLINK/PLINK2	Multiple methods: Cochran-Armitage trend test, Fisher’s exact test, genotypic tests (general, dominant, and recessive models), and Cochran-Mantel-Haenszeltests for stratified tables and standard linear regressions for quantitative traits	* Calculates identity by descent; calculates complete-linkage hierarchical clustering to assess population stratification, OR can use a data-reduction technique of classical MDS to produce a k-dimensional representation of any substructure. * You need to remove closely related individuals	Plink Tutorial Plink2 Tutorial	(Purcell et al. (2007))
LEA (R package)	Latent factor mixed model Lfmm: MCMC Lfmm2: least-squares estimates (for large SNP datasets)	* Uses latent factors to estimate values of K * does not account for closely related individuals–our suggestion would be to remove these individuals from the dataset	Bay Lab Tutorial *Note that phenotypic traits must be in the env format (substitute environmental data for phenotypic traits)	(Frichot et al. (2013)) (Frichot and François (2015)) (François et al. (2016))
fastGWA	mixed linear model	* uses SNP-derived PCs to correct for the effect due to population stratification * estimates relatedness using a genetic relationship matrix (GRM) between pairs of individuals from a set of SNPs	Yang Lab Tutorial	(Jiang et al. (2019))
SNPTEST (improved imputation method)	Options for both frequentist linear models or bayesian models	* uses eigenvectors from a PCA analysis to code for unobserved population structure and included as a covariate in the model * does not account for closely related individuals–our suggestion would be to remove these individuals from the dataset	SNPTest Tutorial	(Marchini et al. (2007))
EMMAX	restricted maximum likelihoods (REML)	* generates kinship matrix (IBS or BN, BN is preferred) using emmax-kin flag * used for both population stratification and closely related individuals	EMMAX Tutorial	(Kang et al. (2010))

So you found some candidate SNPs…What next?

Great! There are many possible follow-up steps and post-hoc validation to choose from!

You might want to investigate whether loci under selection overlap any genes or are in linkage disequilibrium (LD) with genes, consider using the program LDannot if you have adequate genomic resources (Prunier et al. (2019)). You need an annotated genome, but hot tip: If you don’t have an annotated genome, you can email NCBI to annotate it for you, or you can do it yourself with a program such as Liftoff (more details in see Table 5). The program calculates linkage disequilibrium information in your VCF file using known gene regions from your annotation file to identify which loci under selection are in LD with genes.

If you have identified a long list of candidate SNPs that overlap with genes, you can perform a functional enrichment test using Gene Ontology terms (GO; a standardized database of the biological process, molecular function and cellular component of a gene) using R packages such as REVIGO or TopGO. Alternatively, if you have identified a small number of SNPs that overlap with genes (>10 or less; note that this exact number will be subjective), you can look at the functions of just those set of genes, either from the annotated genome itself, or blasting the sequence on NCBI.

Other types of questions that can be answered are the rate of genetic load in candidate SNPs. This can be explored by investigating the occurrence of nonsynonymous SNPs (a variant that changes amino acid in the protein) and the ratio of nonsynonymous vs synonymous substitutions (dN/dS). The increased ratio of dN/dS substitutions are hypothesized to be increased evidence for their relevance to adaptation (Rellstab et al. (2015)).

Another growing area of research is to investigate the genomic vulnerability of the species, which uses identified outlier loci to estimate the magnitude of adaptation required to keep up with predicted changes in climate (Capblancq et al. (2020)). The R package LEA can compute values of genetic offset statistics based on new or predicted environments (see the functions genetic.gap, genetic.offset).

TABLE 5: Annotating your candidate SNPs under selection

A short list of a few resources for annotating your candidate SNPs under selection. This is not an exhaustive list!. We also provide a link to public tutorials for each method.

Program	Program Function	Tutorial	Reference
topGO	Functional enrichment	TopGO Tutorial	Sayols (2023)
REVIGO / rrvgo	Functional enrichment	GUI Web App, rrvgo Manual	Supek et al. (2011)
LD-annot	Find overlap of SNPs with genes	GitHub Repository	Prunier et al. (2019)
Liftoff	Annotate your genome	GitHub Repository	Shumate and Salzberg (2021)

References

Ahrens, Collin W, Paul D Rymer, Adam Stow, Jason Bragg, Shannon Dillon, Kate DL Umbers, and Rachael Y Dudaniec. 2018. “The Search for Loci Under Selection: Trends, Biases and Progress.” Molecular Ecology 27 (6): 1342–56.

Alvarado, Allison H, Christen M Bossu, Ryan J Harrigan, Rachael A Bay, Allison RP Nelson, Thomas B Smith, and Kristen C Ruegg. 2022. “Genotype–Environment Associations Across Spatial Scales Reveal the Importance of Putative Adaptive Genetic Variation in Divergence.” Evolutionary Applications 15 (9): 1390–1407.

Assis, Jorge, Lennert Tyberghein, Samuel Bosch, Heroen Verbruggen, Ester A Serrão, and Olivier De Clerck. 2018. “Bio-ORACLE V2. 0: Extending Marine Data Layers for Bioclimatic Modelling.” Global Ecology and Biogeography 27 (3): 277–84.

Beaumont, Mark A, and Richard A Nichols. 1996. “Evaluating Loci for Use in the Genetic Analysis of Population Structure.” Proceedings of the Royal Society of London. Series B: Biological Sciences 263 (1377): 1619–26.

Bonhomme, Maxime, Claude Chevalet, Bertrand Servin, Simon Boitard, Jihad Abdallah, Sarah Blott, and Magali SanCristobal. 2010. “Detecting Selection in Population Trees: The Lewontin and Krakauer Test Extended.” Genetics 186 (1): 241–62.

Booker, Tom R, Sam Yeaman, James R Whiting, and Michael C Whitlock. 2024. “The WZA: A Window-Based Method for Characterizing Genotype–Environment Associations.” Molecular Ecology Resources 24 (2): e13768.

Brassington, Gary B, Tim Pugh, Claire Spillman, Eric Schulz, Helen Beggs, Andreas Schiller, and Peter R Oke. 2007. “BLUElink> Development of Operational Oceanography and Servicing in Australia.” Journal of Research and Practice in Information Technology 39 (2): 151–64.

Breiman, Leo. 2001. “Random Forests.” Machine Learning 45: 5–32.

Brennan, Reid S, James A DeMayo, Hans G Dam, Michael Finiguerra, Hannes Baumann, Vince Buffalo, and Melissa H Pespeni. 2022. “Experimental Evolution Reveals the Synergistic Genomic Mechanisms of Adaptation to Ocean Warming and Acidification in a Marine Copepod.” Proceedings of the National Academy of Sciences 119 (38): e2201521119.

Brickman, Dave, Michael A Alexander, Andrew Pershing, James D Scott, and Zeliang Wang. 2021. “Projections of Physical Conditions in the Gulf of Maine in 2050.” Elem Sci Anth 9 (1): 00055.

Buffalo, Vince, and Graham Coop. 2020. “Estimating the Genome-Wide Contribution of Selection to Temporal Allele Frequency Change.” Proceedings of the National Academy of Sciences 117 (34): 20672–80.

Capblancq, Thibaut, Matthew C Fitzpatrick, Rachael A Bay, Moises Exposito-Alonso, and Stephen R Keller. 2020. “Genomic Prediction of (Mal) Adaptation Across Current and Future Climatic Landscapes.” Annual Review of Ecology, Evolution, and Systematics 51: 245–69.

Capblancq, Thibaut, and Brenna R Forester. 2021. “Redundancy Analysis: A Swiss Army Knife for Landscape Genomics.” Methods in Ecology and Evolution 12 (12): 2298–2309.

Capblancq, Thibaut, Keurcien Luu, Michael GB Blum, and Eric Bazin. 2018. “Evaluation of Redundancy Analysis to Identify Signatures of Local Adaptation.” Molecular Ecology Resources 18 (6): 1223–33.

Caye, Kevin, Basile Jumentier, Johanna Lepeule, and Olivier François. 2019. “LFMM 2: Fast and Accurate Inference of Gene-Environment Associations in Genome-Wide Studies.” Molecular Biology and Evolution 36 (4): 852–60.

Chamberlain, MA, PR Oke, GB Brassington, P Sandery, P Divakaran, and RAS Fiedler. 2021. “Multiscale Data Assimilation in the Bluelink Ocean Reanalysis (BRAN).” Ocean Modelling 166: 101849.

Coop, Graham, David Witonsky, Anna Di Rienzo, and Jonathan K Pritchard. 2010. “Using Environmental Correlations to Identify Loci Underlying Local Adaptation.” Genetics 185 (4): 1411–23.

Dauphin, Benjamin, Christian Rellstab, Rafael O Wüest, Dirk N Karger, Rolf Holderegger, Felix Gugerli, and Stéphanie Manel. 2023. “Re-Thinking the Environment in Landscape Genomics.” Trends in Ecology & Evolution 38 (3): 261–74.

De Villemereuil, Pierre, and Oscar E Gaggiotti. 2015. “A New FST-Based Method to Uncover Local Adaptation Using Environmental Variables.” Methods in Ecology and Evolution 6 (11): 1248–58.

Duforet-Frebourg, Nicolas, Lucie M Gattepaille, Michael GB Blum, and Mattias Jakobsson. 2015. “HaploPOP: A Software That Improves Population Assignment by Combining Markers into Haplotypes.” BMC Bioinformatics 16: 1–8.

Ellis, Nick, Stephen J Smith, and C Roland Pitcher. 2012. “Gradient Forests: Calculating Importance Gradients on Physical Predictors.” Ecology 93 (1): 156–68.

Enbody, Erik D, Ashley T Sendell-Price, C Grace Sprehn, Carl-Johan Rubin, Peter M Visscher, B Rosemary Grant, Peter R Grant, and Leif Andersson. 2023. “Community-Wide Genome Sequencing Reveals 30 Years of Darwin’s Finch Evolution.” Science 381 (6665): eadf6218.

Ferrier, Simon, Glenn Manion, Jane Elith, and Karen Richardson. 2007. “Using Generalized Dissimilarity Modelling to Analyse and Predict Patterns of Beta Diversity in Regional Biodiversity Assessment.” Diversity and Distributions 13 (3): 252–64.

Fitzpatrick, Matthew C, Vikram E Chhatre, Raju Y Soolanayakanahally, and Stephen R Keller. 2021. “Experimental Support for Genomic Prediction of Climate Maladaptation Using the Machine Learning Approach Gradient Forests.” Molecular Ecology Resources 21 (8): 2749–65.

Fitzpatrick, Matthew C, and Stephen R Keller. 2015. “Ecological Genomics Meets Community-Level Modelling of Biodiversity: Mapping the Genomic Landscape of Current and Future Environmental Adaptation.” Ecology Letters 18 (1): 1–16.

Foll, Matthieu, and Oscar Gaggiotti. 2008. “A Genome-Scan Method to Identify Selected Loci Appropriate for Both Dominant and Codominant Markers: A Bayesian Perspective.” Genetics 180 (2): 977–93.

Forester, Brenna R, Amanda S Cicchino, Alisha A Shah, Austin B Mudd, Eric C Anderson, Jessen V Bredeson, Andrew J Crawford, et al. 2025. “Population Genomics Reveals Local Adaptation Related to Temperature Variation in Two Stream Frog Species: Implications for Vulnerability to Climate Warming.” Molecular Ecology, e17651.

Forester, Brenna R, Matthew R Jones, Stéphane Joost, Erin L Landguth, and Jesse R Lasky. 2016. “Detecting Spatial Genetic Signatures of Local Adaptation in Heterogeneous Landscapes.” Molecular Ecology 25 (1): 104–20.

François, Olivier, Helena Martins, Kevin Caye, and Sean D Schoville. 2016. “Controlling False Discoveries in Genome Scans for Selection.” Molecular Ecology 25 (2): 454–69.

Frichot, Eric, and Olivier François. 2015. “LEA: An r Package for Landscape and Ecological Association Studies.” Methods in Ecology and Evolution 6 (8): 925–29.

Frichot, Eric, Sean D Schoville, Guillaume Bouchard, and Olivier François. 2013. “Testing for Associations Between Loci and Environmental Gradients Using Latent Factor Mixed Models.” Molecular Biology and Evolution 30 (7): 1687–99.

Fuller, Zachary L, Veronique JL Mocellin, Luke A Morris, Neal Cantin, Jihanne Shepherd, Luke Sarre, Julie Peng, et al. 2020. “Population Genetics of the Coral Acropora Millepora: Toward Genomic Prediction of Bleaching.” Science 369 (6501): eaba4674.

Gautier, Mathieu. 2015. “Genome-Wide Scan for Adaptive Divergence and Association with Population-Specific Covariates.” Genetics 201 (4): 1555–79.

Griffiths, Joanna S, Kendall Smith, and Andrew Whitehead. 2025. “Seascape Genomics of Red Abalone: Limited Range-Wide Population Structure and Evidence for Local Adaptation.” Molecular Ecology, e17650.

Hoban, Sean, Catriona D Campbell, Jessica M da Silva, Robert Ekblom, W Chris Funk, Brittany A Garner, José A Godoy, Francine Kershaw, Anna J MacDonald, and Joachim Mergeay. 2021. “Genetic Diversity Is Considered Important but Interpreted Narrowly in Country Reports to the Convention on Biological Diversity: Current Actions and Indicators Are Insufficient.” Biological Conservation 261: 109233.

Hoban, Sean, Joanna L Kelley, Katie E Lotterhos, Michael F Antolin, Gideon Bradburd, David B Lowry, Mary L Poss, Laura K Reed, Andrew Storfer, and Michael C Whitlock. 2016. “Finding the Genomic Basis of Local Adaptation: Pitfalls, Practical Solutions, and Future Directions.” The American Naturalist 188 (4): 379–97.

Jean-Michel, Lellouche, Greiner Eric, Bourdallé-Badie Romain, Garric Gilles, Melet Angélique, Drévillon Marie, Bricaud Clément, et al. 2021. “The Copernicus Global 1/12 Oceanic and Sea Ice GLORYS12 Reanalysis.” Frontiers in Earth Science 9: 698876.

Jiang, Longda, Zhili Zheng, Ting Qi, Kathryn E Kemper, Naomi R Wray, Peter M Visscher, and Jian Yang. 2019. “A Resource-Efficient Tool for Mixed Model Association Analysis of Large-Scale Data.” Nature Genetics 51 (12): 1749–55.

Jørsboe, Emil, and Anders Albrechtsen. 2022. “Efficient Approaches for Large-Scale GWAS with Genotype Uncertainty.” G3 12 (1): jkab385.

Kang, Hyun Min, Jae Hoon Sul, Susan K Service, Noah A Zaitlen, Sit-yee Kong, Nelson B Freimer, Chiara Sabatti, and Eleazar Eskin. 2010. “Variance Component Model to Account for Sample Structure in Genome-Wide Association Studies.” Nature Genetics 42 (4): 348–54.

Kijas, James, Sean McWilliam, Marina Naval Sanchez, Peter Kube, Harry King, Bradley Evans, Torfinn Nome, Sigbjørn Lien, and Klara Verbyla. 2018. “Evolution of Sex Determination Loci in Atlantic Salmon.” Scientific Reports 8 (1): 5664.

Lellouche, Jean-Michel, Eric Greiner, Olivier Le Galloudec, Gilles Garric, Charly Regnier, Marie Drevillon, Mounir Benkiran, et al. 2018. “Recent Updates to the Copernicus Marine Service Global Ocean Monitoring and Forecasting Real-Time 1/12∘ High-Resolution System.” Ocean Science 14 (5): 1093–1126.

Lotterhos, Katie E. 2019. “The Effect of Neutral Recombination Variation on Genome Scans for Selection.” G3: Genes, Genomes, Genetics 9 (6): 1851–67.

Lotterhos, Katie E, Matthew C Fitzpatrick, and Heath Blackmon. 2022. “Simulation Tests of Methods in Evolution, Ecology, and Systematics: Pitfalls, Progress, and Principles.” Annual Review of Ecology, Evolution, and Systematics 53: 113–36.

Lotterhos, Katie E, and Michael C Whitlock. 2014. “Evaluation of Demographic History and Neutral Parameterization on the Performance of FST Outlier Tests.” Molecular Ecology 23 (9): 2178–92.

———. 2015. “The Relative Power of Genome Scans to Detect Local Adaptation Depends on Sampling Design and Statistical Method.” Molecular Ecology 24 (5): 1031–46.

Lou, Runyang Nicolas, Arne Jacobs, Aryn P Wilder, and Nina Overgaard Therkildsen. 2021. “A Beginner’s Guide to Low-Coverage Whole Genome Sequencing for Population Genomics.” Molecular Ecology 30 (23): 5966–93.

Luu, Keurcien, Eric Bazin, and Michael GB Blum. 2017. “Pcadapt: An r Package to Perform Genome Scans for Selection Based on Principal Component Analysis.” Molecular Ecology Resources 17 (1): 67–77.

Marchini, Jonathan, Bryan Howie, Simon Myers, Gil McVean, and Peter Donnelly. 2007. “A New Multipoint Method for Genome-Wide Association Studies by Imputation of Genotypes.” Nature Genetics 39 (7): 906–13.

Narum, Shawn R, and Jon E Hess. 2011. “Comparison of FST Outlier Tests for SNP Loci Under Selection.” Molecular Ecology Resources 11: 184–94.

Nielsen, Erica S, Romina Henriques, Maria Beger, and Sophie von der Heyden. 2021. “Distinct Interspecific and Intraspecific Vulnerability of Coastal Species to Global Change.” Global Change Biology 27 (15): 3415–31.

Pérez-Figueroa, A, MJ Garcı́a-Pereira, M Saura, E Rolán-Alvarez, and A Caballero. 2010. “Comparing Three Different Methods to Detect Selective Loci Using Dominant Markers.” Journal of Evolutionary Biology 23 (10): 2267–76.

Pickrell, Joseph, and Jonathan Pritchard. 2012. “Inference of Population Splits and Mixtures from Genome-Wide Allele Frequency Data.” Nature Precedings, 1–1.

Pinsky, Malin L. 2021. “Diversification Spins a Heatwave Safety Net for Fisheries.” Proceedings of the National Academy of Sciences 118 (3): e2024412118.

Prunier, Julien, Audrey Lemaçon, Alexandre Bastien, Mohsen Jafarikia, Ilga Porth, Claude Robert, and Arnaud Droit. 2019. “LD-Annot: A Bioinformatics Tool to Automatically Provide Candidate SNPs with Annotations for Genetically Linked Genes.” Frontiers in Genetics 10: 1192.

Purcell, Shaun, Benjamin Neale, Kathe Todd-Brown, Lori Thomas, Manuel AR Ferreira, David Bender, Julian Maller, et al. 2007. “PLINK: A Tool Set for Whole-Genome Association and Population-Based Linkage Analyses.” The American Journal of Human Genetics 81 (3): 559–75.

Rellstab, Christian, Felix Gugerli, Andrew J Eckert, Angela M Hancock, and Rolf Holderegger. 2015. “A Practical Guide to Environmental Association Analysis in Landscape Genomics.” Molecular Ecology 24 (17): 4348–70.

Riginos, Cynthia, Eric D Crandall, Libby Liggins, Pim Bongaerts, and Eric A Treml. 2016. “Navigating the Currents of Seascape Genomics: How Spatial Analyses Can Augment Population Genomic Studies.” Current Zoology 62 (6): 581–601.

Robertson, Alan. 1975. “Gene Frequency Distributions as a Test of Selective Neutrality.” Genetics 81 (4): 775–85.

Rumberger, Camille, Madison Armstrong, Martin Kim, Josue Melendez, Raquel Ponce, Melissa DeBiasse, Serena Caplins, and Rachael Bay. 2023. “Selection over Small and Large Spatial Scales in the Face of High Gene Flow.” Authorea Preprints.

Sayols, Sergi. 2023. “Rrvgo: A Bioconductor Package for Interpreting Lists of Gene Ontology Terms.” Micropublication Biology 2023.

Schlötterer, Christian, Raymond Tobler, Robert Kofler, and Viola Nolte. 2014. “Sequencing Pools of Individuals—Mining Genome-Wide Polymorphism Data Without Big Funding.” Nature Reviews Genetics 15 (11): 749–63.

Selmoni, Oliver, Estelle Rochat, Gael Lecellier, Veronique Berteaux-Lecellier, and Stéphane Joost. 2020. “Seascape Genomics as a New Tool to Empower Coral Reef Conservation Strategies: An Example on North-Western Pacific Acropora Digitifera.” Evolutionary Applications 13 (8): 1923–38.

Shumate, Alaina, and Steven L Salzberg. 2021. “Liftoff: Accurate Mapping of Gene Annotations.” Bioinformatics 37 (12): 1639–43.

Silliman, Katherine. 2019. “Population Structure, Genetic Connectivity, and Adaptation in the Olympia Oyster (Ostrea Lurida) Along the West Coast of North America.” Evolutionary Applications 12 (5): 923–39.

Supek, Fran, Matko Bošnjak, Nives Škunca, and Tomislav Šmuc. 2011. “REVIGO Summarizes and Visualizes Long Lists of Gene Ontology Terms.” PloS One 6 (7): e21800.

Tam, Vivian, Nikunj Patel, Michelle Turcotte, Yohan Bossé, Guillaume Paré, and David Meyre. 2019. “Benefits and Limitations of Genome-Wide Association Studies.” Nature Reviews Genetics 20 (8): 467–84.

Uffelmann, Emil, Qin Qin Huang, Nchangwi Syntia Munung, Jantina De Vries, Yukinori Okada, Alicia R Martin, Hilary C Martin, Tuuli Lappalainen, and Danielle Posthuma. 2021. “Genome-Wide Association Studies.” Nature Reviews Methods Primers 1 (1): 59.

Van Wyngaarden, Mallory, Paul VR Snelgrove, Claudio DiBacco, Lorraine C Hamilton, Naiara Rodrı́guez-Ezpeleta, Nicholas W Jeffery, Ryan RE Stanley, and Ian R Bradbury. 2017. “Identifying Patterns of Dispersal, Connectivity and Selection in the Sea Scallop, Placopecten Magellanicus, Using RAD Seq-Derived SNP s.” Evolutionary Applications 10 (1): 102–17.

Wang, Shuangqiang, Ellen Kenchington, Zeliang Wang, and Andrew J Davies. 2021. “Life in the Fast Lane: Modeling the Fate of Glass Sponge Larvae in the Gulf Stream.” Frontiers in Marine Science 8: 701218.

Whitlock, Michael C, and Katie E Lotterhos. 2015. “Reliable Detection of Loci Responsible for Local Adaptation: Inference of a Null Model Through Trimming the Distribution of f ST.” The American Naturalist 186 (S1): S24–36.

Wooldridge, Brock, Chloé Orland, Erik Enbody, Merly Escalona, Cade Mirchandani, Russell Corbett-Detig, Joshua D Kapp, et al. 2024. “Limited Genomic Signatures of Population Collapse in the Critically Endangered Black Abalone (Haliotis Cracherodii).” Molecular Ecology, e17362.

Wright, Sewall. 1949. “Population Structure in Evolution.” Proceedings of the American Philosophical Society 93 (6): 471–78.

Xuereb, Amanda, Laura Benestan, Éric Normandeau, Rémi M Daigle, Janelle MR Curtis, Louis Bernatchez, and Marie-Josée Fortin. 2018. “Asymmetric Oceanographic Processes Mediate Connectivity and Population Genetic Structure, as Revealed by RAD Seq, in a Highly Dispersive Marine Invertebrate (Parastichopus Californicus).” Molecular Ecology 27 (10): 2347–64.

Yanez, Jose M, Agustin Barria, Maria E Lopez, Thomas Moen, Baltasar F Garcia, Grazyella M Yoshida, and Peng Xu. 2023. “Genome-Wide Association and Genomic Selection in Aquaculture.” Reviews in Aquaculture 15 (2): 645–75.

Zhou, Xiang, and Matthew Stephens. 2012. “Genome-Wide Efficient Mixed-Model Analysis for Association Studies.” Nature Genetics 44 (7): 821–24.

———. 2014. “Efficient Multivariate Linear Mixed Model Algorithms for Genome-Wide Association Studies.” Nature Methods 11 (4): 407–9.