Regarding theF-word: the effects of dataFilteringon inferred genotype-environment associations

Abstract

Genotype-environment association (GEA) methods have become part of the standard landscape genomics toolkit, yet, we know little about how to filter genotype-by-sequencing data to provide robust inferences for environmental adaptation. In many cases, default filtering thresholds for minor allele frequency and missing data are applied regardless of sample size, having unknown impacts on the results. These effects could be amplified in downstream predictions, including management strategies. Here, we investigate the effects of filtering on GEA results and the potential implications for adaptation to environment. Using empirical and simulated datasets derived from two widespread tree species to assess the effects of filtering on GEA outputs. Critically, we find that the level of filtering of missing data and minor allele frequency affect the identification of true positives. Even slight adjustments to these thresholds can change the rate of true positive detection. Using conservative thresholds for missing data and minor allele frequency substantially reduces the size of the dataset, lessening the power to detect adaptive variants (i.e. simulated true positives) with strong and weak strength of selections. Regardless, strength of selection was a good predictor for GEA detection, but even SNPs under strong selection went undetected. We further show that filtering can significantly impact the predictions of adaptive capacity of species in downstream analyses. We make several recommendations regarding filtering for GEA methods. Ultimately, there is no filtering panacea, but some choices are better than others, depending largely on the study system, availability of genomic resources, and desired objectives of the study.

Scale‐dependent signatures of local adaptation in a foundation tree species

“…Interestingly, no overlap in outlier loci between large and small spatial scales was found in a study of the Amazonian forest tree Eperua falcata (Brousseau et al, 2020), suggestive of the presence of multiple adaptation optima occurring through different genetic architectures. Similarly to our study, Brousseau et al (2020) found a substantial number of SNPs in nongenic regions of the genome that were associated with environmental variation, highlighting the ways that experimental designs including multiple spatial scales and annotated reference genomes can increase the potential for identifying false positives (Ahrens et al, 2021). A study of great tits ( Parus major ) in France also found that spatial scale was an important determinant of the significance of urbanisation on genetic differentiation (Perrier et al, 2018), and studies of salmonid fishes have shown that geographically distant populations experience markedly different selection regimes compared to more proximate populations (Fraser et al, 2011), which is likely to give rise to different genomic architectures relating to adaptation.…”

“…A total of 51 SNPs were under putative selection at both spatial scales, of which 39 were genic or near‐genic, associated with 28 individual genes. No SNP was identified by both LFMM and RDA at both spatial scales, highlighting the substantially different results that can be obtained depending on the choice of GEA method and spatial scale of sampling (Ahrens et al, 2021). Further, our hypothesis that fundamental differences between spatial scales will result in stronger signals of adaptation in the range‐wide spatial scale was correct.…”

Closely Related Tree Species with Overlapping Ranges Exhibit Divergent Adaptation to Climate

“…Table S1 & S2) using the Moran.I function in the package APE (Paradis & Schliep, 2019) to determine the effective sample size (ESS) of each population based on environmental independence. It is worth noting that the eight climatic variables used on both species were not entirely independent, but showed low correlations within one species or the other, though values of r up to 0.7 do not increase the number of false positives detected (Ahrens et al, 2021).…”

“…Multiple testing can be done using Bonferroni corrections applied to the calibrated p-values. We applied a significance value of (a = 0.001 since lower thresholds were found to be more variable and returned greater numbers of false positives (Ahrens et al, 2021). These thresholds were applied to each of the eight climate variables.…”

Commonly used Hardy-Weinberg equilibrium filtering schemes impact population structure inferences using RADseq data

“…Linck & Battey (2019) showed that minor allele frequency (MAF) filtering of datasets may be problematic since it alters the site frequency spectrum (SFS) across loci according to their rate of missingness. Additional recent work has revealed that both variant call rate and MAF can affect population genetic inferences and genotype-environment association studies (Ahrens et al, 2021; Selechnik et al, 2020). In Table 1, we summarise filtering approaches that are commonly applied to RADseq data, the reasons for their usage, and how they can affect population genetic inference.…”