A Multifaceted Approach Reveals Complex Genomic Mediation of White‐Nose Syndrome Adaptive Response in the Little Brown Bat ( Myotis lucifugus )

ABSTRACT

Novel pathogens have become a major challenge faced by wildlife in the Anthropocene. White‐nose syndrome (WNS) has decimated bat populations across North America over the last two decades. Demographic and physiological evidence of adaptation in one heavily affected species, Myotis lucifugus , has prompted multiple attempts to delineate the genomic underpinnings, but these studies show little congruence in their findings. This may be due, in part, to the limitations of the genomic resources utilized and/or analytical approaches employed. Here, we performed high‐coverage whole‐genome resequencing of M. lucifugus sampled prior to (n = 29), and 10 years after the local arrival of WNS (n = 30), aligned to a new reference genome to identify signatures of selection associated with adaptive responses. Using 41.9 million SNPs, we implemented a combination of hard and soft sweep detection analyses, leading to the discovery of 382 genes with robust evidence of selection. Of these, 202 (49.9%) were associated with enriched gene ontology (GO) terms, many of which were tied to neuron development, organization and function. Further, approximately half (107) of genes associated with enriched GO terms interact with genes identified by previous studies. Our findings suggest reduced susceptibility to WNS is mediated through highly complex, polygenic mechanisms. Further, we demonstrate there are far more connections among WNS selection study results than previously recognized. We believe that the methods employed by our study illustrate a need for a paradigm shift in non‐model selection studies and further highlight the value of genomics as a tool for conservation management.

1. Introduction

The Anthropocene continues to introduce new challenges to wildlife, one of the most daunting being the global spread of pathogens (Cunningham et al. 2017). How populations respond to novel pathogens dictates short‐ and long‐term viability and, in turn, the necessity for conservation intervention. A multitude of approaches have been used to assess populations' adaptive responses to a given pathogen, including assessments of shifts in behaviour and physiology (Lilley et al. 2016), immunological profiles (Hecht‐Höger et al. 2020) and gene expression patterns (Davy et al. 2020). In addition, advances in genomic sequencing and analytical tools offer the opportunity to investigate rapid adaptation to disease through identifying signatures of selection and revealing the biological mechanisms of reduced susceptibility. Understanding the basis of adaptive disease responses could translate to efficacious management practices to rescue populations in the throes of epidemic or develop prophylactics for naïve populations.

Bats (order: Chiroptera) are particularly susceptible to anthropogenic pressures due to their low reproductive rates (Barclay et al. 2004), complex seasonal habitat needs (Gili et al. 2020) and longevity (Munshi‐South and Wilkinson 2010), resulting in 18% of global bat species currently being considered threatened (IUCN 2025). At present, one of the largest threats to many North American species is white‐nose syndrome (WNS), a disease caused by the fungal pathogen Pseudogymnoascus destructans (Pd; Lorch et al. 2011), a cold‐loving fungus native to Eurasia (Leopardi et al. 2015). WNS was first detected in Myotis lucifugus (common name the little brown bat) in New York, USA in 2006, noted by distinct white fungal growth on the muzzles, ears and wings of hibernating bats (Blehert et al. 2009). Today, WNS has gradually spread across North America impacting a growing number of bat species (Hoyt et al. 2021), most recently reaching California, USA in 2024 (California Department of Fish and Wildlife 2024). Interestingly, Nearctic bat species present highly variable responses to Pd ranging from seemingly tolerant (i.e., limited pathogen load and no recorded mortality events; e.g., Corynorhinus rafinesqui) to extremely susceptible (i.e., excessive pathogen load and high mortality rates; e.g., Myotis septentrionalis ; Hoyt et al. 2021).

WNS pathology is complex—proliferation of fungal conidia in the epidermis causes cupping erosions on epidermal tissue (Reeder et al. 2012) leading to increased evaporative water loss (Warnecke et al. 2013), systemic inflammation (Field et al. 2015, 2018; Lilley et al. 2019) and frequent arousal from torpor resulting in energy depletion during hibernation (Reeder et al. 2012), which culminates in a pathophysiological cascade of responses that irrevocably disrupt homeostasis (Hoyt et al. 2021; Warnecke et al. 2013). Despite the high morbidity of WNS in M. lucifugus (> 90% mortality in many hibernacula; Hoyt et al. 2021; Langwig et al. 2012), a number of populations have avoided extirpation (Cheng et al. 2021; Langwig et al. 2017), some of which contain individuals that have survived the entirety of Pd presence (Kurta et al. 2020). Following the initial epidemic, population numbers are recovering in multiple hibernacula that were once significantly impacted by WNS despite Pd persistence (Frick et al. 2017; Langwig et al. 2015). Moreover, wildlife managers have noted an absence of classic manifestations of WNS in those hibernacula with the longest history of Pd exposure (G. G. Turner, Pennsylvania Game Commission, personal communication). Together, these observations indicate an adaptive response has emerged following the strong selective pressure WNS has imposed.

Adaptation conferring persistence in the face of continued Pd exposure is multifaceted. Extensive work has revealed a number of behavioural, physiological and biochemical responses to WNS in M. lucifugus , including reversion of hibernation arousal rates (Lilley et al. 2016), shifts in hibernacula microclimate choice (Langwig et al. 2012), increased pre‐hibernation fat deposition (Cheng et al. 2019), downregulation of immune‐gene expression (Kwait et al. 2024), anti‐fungal immune activation (Field et al. 2015), diminished superficial pathogen growth (Cheng et al. 2019; Grimaudo et al. 2022; Langwig et al. 2017) and reduced disease severity (Gagnon et al. 2025; Grimaudo et al. 2022). While some adaptations indicate signs of disease tolerance, others imply persistence mediated through resistance. Given these mechanisms are not mutually exclusive (Restif and Koella 2004) and the variety of adaptive responses to WNS, it is likely that both tolerance and resistance mediate survival to some degree. To support recovering populations and aid those in the midst of epidemic, several infection mitigation and eradication efforts have been developed, namely antifungal volatile organic compound application (Cornelison et al. 2014), microbiome inoculation (Hoyt et al. 2015), microclimate manipulation (Turner et al. 2022) and immunization (Rocke et al. 2019), but have thus far shown mixed results. However, if the biological mechanism(s) of WNS resistance and/or tolerance could be identified, the efficacy of current management strategies may be markedly improved or lead to the development of new approaches for curbing or even halting epidemics.

To date, studies of adaptive selection following the arrival of WNS have shown little consensus regarding the gene variants under selection. One explanation may be the limitations of the genomic resources utilized, including a M. lucifugus reference genome generated using short‐read sequencing technology (Myoluc2.0; GenkBank accession: GCA_000147115.1) alongside either targeted capture sequencing (Donaldson et al. 2017), reduced representation sequencing (Auteri and Knowles 2020), or low‐coverage whole‐genome resequencing (WGR; Gignoux‐Wolfsohn et al. 2021; Lilley et al. 2020), limiting data resolution and accuracy (Stephens and Iyer 2018). In addition, these studies in large part utilized hard sweep detection analyses, generally disregarding the potential for a soft sweep model of selection, and lacked assessments of polygenic selection. Moreover, no quantitative investigations into connections among study results have been performed.

The approaches of previous studies follow a trend of how selection studies have been performed in non‐model systems, focusing largely on hard sweep analyses (Haasl and Payseur 2016). Advances in genomic analytical methods now facilitate genome scans capable of detecting other modes of selection, including soft sweeps (Vatsiou et al. 2016). Unlike hard sweeps, where selection acts on a single new mutation, soft sweeps occur when standing variation becomes beneficial following a novel selective pressure (Hermisson and Pennings 2005). Hence, multiple haplotypes increase in frequency, resulting in patterns of linkage and haplotype diversity disparate from those produced by selection on a single variant (Pennings and Hermisson 2006). Studies also appear to emphasize methods aimed at identifying oligogenic (i.e., adaptation mediated via few genes; e.g., Auteri and Knowles 2020; de Greef et al. 2022) rather than polygenic selection. Though methods identifying shared function among candidate genes (e.g., gene ontology enrichment) within studies are common, quantitative approaches for revealing biological interactions among genes detected across multiple studies, while standard in human disease studies (Zhang et al. 2020), are rare in non‐model systems. Expanding standard practices for detecting selection is essential, particularly for systems of rapid adaptation to disease, as soft sweeps are likely the dominant mode of selection under short timescales (Messer and Petrov 2013) and adaptation to pathogens is often mediated through polygenic mechanisms (Barghi et al. 2020). To date, no study has performed high‐coverage whole‐genome resequencing, comprehensive soft sweep analyses, or quantitative approaches to detect polygenic selection to examine the genomic signatures of WNS survival in bats.

In this study we improve on previous attempts to delineate the genomic response of M. lucifugus populations to WNS by utilizing high‐quality genomic resources in conjunction with a comprehensive approach for detecting signatures of selection. Specifically, we employed high‐coverage WGR of both naïve and post‐epidemic M. lucifugus exhibiting signs of reduced WNS susceptibility (Figure 1A) alongside a new chromosome‐level reference genome assembly to (1) assess genome‐wide changes in variation, (2) identify genes under selection in persisting populations, (3) interpret possible biological mechanisms of adaptive disease response, (4) compare results back to those of previous studies and (5) exemplify the utility of employing broader genomic approaches for selection studies. We were prepared to find signatures of selection indicating immune response as a major mechanism of WNS adaptation similar to previous studies. However, given the robustness of the genomic resources utilized here, we anticipated the discovery of new genes of importance. By incorporating the findings of previous studies without restricting our own, we expect this study will reveal unprecedented insights into the genomic underpinnings of WNS survival as well as illustrate the utility of expanded approaches to detecting selection in populations of conservation concern.

FIGURE 1.

Population sampling and whole‐genome population structure of white‐nose syndrome affected Myotis lucifugus colonies. (A) A map depicting the location, number per site, and timing of samples collected for this study. Population names with the suffix ‘PRE’ refer to samples predating the local arrival of white‐nose syndrome while ‘POST’ indicates samples collected ~10 years after. (B) A principal component analysis (PCA) performed using a subset (n = 346,939) of the final filtered SNPs after removing one individual per closely related pair (n = 2; see Figure S4). The dataset was thinned to retain SNPs at least 5 Kbp apart to reduce the influence of linkage. Ellipses represent 95% confidence intervals and the percent variation explained by each principal component is specified in parentheses.

2. Materials and Methods

2.1. Population Sampling and Sequencing

We conducted WGR on M. lucifugus individuals from eastern United States (New York [NY] and Pennsylvania [PA]) colonies sampled preceding (n = 29) and ~10 years following (n = 30) the arrival of WNS (Figure 1A). Samples are referenced both by their original population designations (i.e., ‘NYPRE’, ‘NYPOST’, ‘PAPRE’ and ‘PAPOST’; see Table S1) as well as by their collective sampling timepoint, with all population designations containing ‘PRE’ collectively referred to as ‘preWNS’ and all population designations containing ‘POST’ collectively referred to as ‘postWNS’. We selected these 59 samples from a larger collection generated by Lilley et al. (2020) based on DNA quality, library size distribution and even representation of each sampling location and timepoint. For sample collection and DNA extraction details see Lilley et al. (2020). Prior to library construction, we assessed DNA quality using an Agilent Bioanalyser (Agilent, Santa Clara, CA, USA). We generated two separate WGR libraries: one containing 8 samples designated for high‐coverage sequencing (HC) and another containing the remaining 51 samples selected for shallow coverage sequencing (SC). Libraries were generated using the NEBnext II Ultra FS DNA library prep kit for Illumina (New England Biolabs, Ipswich, MA, USA) following the < 100 ng input protocol. First, we fragmented DNA samples targeting lengths of 600 bp and uniquely barcoded by ligating combinatorial dual index adapters using NEBnext Multiplex Oligos for Illumina (New England Biolabs, Ipswich, MA, USA) at 1.5 μM concentration. Individual libraries were amplified via PCR as described in the FS DNA library prep kit protocol followed by clean‐up and size selection using SPRIselect beads (Beckman Coulter Life Sciences, Indianapolis, IN, USA). We quantified library concentration using a SpectraMax M3 Absorbance Reader (Molecular Devices, San Jose, CA, USA) and AccuBlue Broad Range dsDNA Quantitation Kit (Biotium Inc., Fremont, CA, USA) and assessed fragment size using a QIAxcel Advanced (QIAGEN, Venlo, Netherlands). Final libraries were then pooled for sequencing.

The UC Davis Genome Center DNA Technologies and Expression Analysis Core sequenced the final libraries according to their standard practices for WGR library sequencing under the NIH Shared Instrumentation Grant (no. 1S10OD010786‐01). Briefly, fragment size distribution of each library was verified using the LabChip GX Touch system (PerkinElmer, Waltham, MA, USA). Libraries were quantified via Qubit fluorometer (LifeTechnologies, Carlsbad, CA, USA) then pooled in equimolar ratios. Primer dimers were removed from both pools by SPRI‐bead purification following the manufacturer's recommendations (Ampure XP beads; Beckman Coulter, Brea, CA, USA) and pools were quantified using the Kapa Library Quant kit (Kapa Biosystems‐Roche, Basel, Switzerland) on a QuantStudio 5 real‐time PCR system (Applied Biosystems, Foster City, CA, USA). Pooled libraries were sequenced on an Illumina NovaSeq 6000 S4 flowcell (Illumina, San Diego, CA, USA) using 2 × 150 paired‐end reads; the HC library was sequenced on one lane while the SC library was sequenced across three lanes. Finally, reads were demultiplexed and assigned to individual samples. Demultiplexed raw sequencing reads are available through NCBI (BioProject PRJNA1353610, accession nos. SAMN52933055–SAMN52933113).

2.2. Read Alignment and SNP Discovery

Raw sequencing reads were cleaned and trimmed using htstream v1.3.3 (https://github.com/s4hts/HTStream). Preprocessing steps included screening out PhiX reads, trimming adapter sequences and low quality end calls, and filtering out reads shorter than 100 bp. Cleaned reads were then aligned using dragmap v1.3.0 (https://github.com/Illumina/DRAGMAP), which has been shown to outperform other commonly used aligners (Betschart et al. 2022), using default settings to a new M. lucifugus reference genome. This assembly was generated using long‐read sequencing paired with proximity ligation (mMyoLuc1, GenBank accession: JAVIYB000000000; see Vazquez et al. (2026) for sequencing, assembly, and annotation details). The greatly improved continuity and completeness over the previous assembly (Figure S1) increased single nucleotide polymorphism (SNP) call reliability, as mapping quality can directly correlate with scaffold length (Stephens and Iyer 2018), as well as expanded selection scan analysis opportunities. We called SNPs using gatk v4.3 (Poplin et al. 2017) then filtered using a combination of GATK and bcftools v1.19 (Danecek et al. 2021). We followed the Broad Institute's hard‐filtering recommendations for non‐model organisms (Broad Institute 2024b), which were quantitatively verified (Figure S2). We further validated the mapping quality (MQ) filter to ensure exclusion of low mappability regions (Figure S3). To be considered for downstream analysis, we required SNPs to fulfil the following parameters: biallelic, present in ≥ 80% of individuals, minor allele count of 5 (i.e., minimum allele frequency of 0.04), read depth between 10× and 200×, DRAGEN mapping quality (MQ) ≥ 166.67, quality by depth (QD) ≥ 2 and symmetric odds ratio (SOR) ≤ 3. We refer to this dataset hereafter as the ‘full SNP dataset’. To limit the influence of linkage for a subset of our analyses, we further filtered the full SNP dataset by thinning to 1 SNP per 5 Kbp, which we hereafter refer to as the ‘thinned SNP dataset’. We performed quality control assessments on the final set of SNPs to evaluate possible influences of depth of coverage on observed heterozygosity and variant allele frequency variability (Hansen et al. 2022; see Supplementary Methods S1). The complete SNP calling pipeline along with execution instructions can be found on GitHub (https://github.com/slcapel/DRAGEN‐GATK4_SNP_calling_pipeline); all other data and code associated with this study can be found on Dryad (https://doi.org/10.5061/dryad.ncjsxkt66).

2.3. Summary Statistics and Population Structure

We assessed genome‐wide patterns of population structure across sampling timepoint and/or geographic location. For each summary statistic, original population designations were considered as well as aggregates of individuals both across sampling time points (i.e., ‘PRE’ and ‘POST’) and geographic origin (i.e., ‘NY’ and ‘PA’). We estimated genome‐wide mean values for several genetic diversity statistics across all population designations using the full SNP dataset. Population mean nucleotide diversity (π) was measured using the per‐site vcftools v0.1.16 (Danecek et al. 2011) estimate. We also calculated global weighted pairwise F _ST based on unfolded sample allele frequencies using ANGSD v0.940 (Korneliussen et al. 2014). See Section 2.4.2 for details on allele polarization methods.

We evaluated genome‐wide population structure using the thinned SNP dataset (n = 346,939) by principal component analysis (PCA) using the r v4.3.1 (R Core Team 2013); packages ade4 (Dray et al. 2020) and factoextra (Kassambara and Mundt 2020). To evaluate the potential confounding factor of high relatedness we calculated kinship of all pairwise individuals with SNPrelate v1.42.0 (Zheng et al. 2012) using the maximum likelihood estimation of identity‐by‐descent approach (Malinsky et al. 2018). Kinship estimates revealed low levels of relatedness among pairs of individuals save for two pairs from the NYPOST population (NYDEC33026_S44–NYDEC33007_S47 and NYDEC33024_S7–NYDEC33023_S46; Figure S4); removal of one individual per related pair from the PCA had little effect on the overall distribution of individuals across PCs.

2.4. Detecting Signatures of Selection

We employed a multifaceted approach to assessing signatures of selection (Figure 2), employing both hard and soft sweep pairwise population detection methods. To investigate the genomic effects of WNS on persisting populations, we centered analyses around disease timepoint comparisons (preWNS–postWNS; i.e., NYPRE–NYPOST, PAPRE–PAPOST, NYPRE–PAPOST, PAPRE–NYPOST and PRE–POST). We reduced the confounding influence of geographic differentiation on signatures of selection by omitting outlier regions that fell within 20 Kbp of geographic outliers (i.e., statistically significant regions identified within strictly geographic comparisons (NYPRE–PAPRE, NYPOST–PAPOST and NY–PA)). Following identification of putative genes under selection (described below), we recognized the breadth of previous work identifying genes involved in WNS response, hereafter referred to as ‘a priori genes of interest’ (Table S2), by incorporating their findings into our own analyses.

FIGURE 2.

Comprehensive methodology for detecting signatures of selection. Depicted in this diagram are the three major phases implemented for detecting and interpreting signatures of selection in this study, including both hard and soft sweep methodologies. In Phase 1, pairwise selection statistics are estimated across the genome in 10 Kbp windows for all disease (preWNS–postWNS) and geographic (New York–Pennsylvania) population comparisons (step 1.1). Significant outlier regions with no confounding effect of geography are identified (step 1.2) as well as the coincident genes (step 1.3). This phase (steps 1.1–1.3) is executed for both hard and soft sweep selection statistics. In Phase 2, genes coincident with outlier regions significant for both statistics (F _ST and XP‐CLR for hard sweeps or Rsb and XP‐CLR for soft sweeps) are identified as under putative selection (step 2.1). Genes with evidence of hard and/or soft sweep selection undergo gene ontology (GO) term enrichment analyses (step 2.2). In Phase 3, genes under putative selection associated with enriched GO terms are combined with genes of interest identified by previous studies (step 3.1) in a gene network analysis to identify gene interactions and interpret broader biochemical pathways involved in adaptive responses to WNS (step 3.2).

2.4.1. Hard Sweep Selection Discovery

To detect hard sweep selection we utilized two statistics: F _ST, estimated using ANGSD v0.940 (Korneliussen et al. 2014), and XP‐CLR, a likelihood‐based multilocus allele frequency differentiation statistic (Chen et al. 2010), to assess genetic differentiation following the arrival of WNS. The F _ST ‐based selection scan takes the classic approach of assuming the majority of loci fall under neutral expectations and that significant right‐handed departures from that approximate Gaussian distribution indicate regions under positive selection. XP‐CLR, on the other hand, models genetic drift under neutrality using Brownian motion, then estimates the degree of departure from neutral expectations based on the length of multilocus allele frequency differentiation. Each statistic was calculated using the full SNP dataset parsed into 10 Kbp sliding windows with 10% overlap. Due to the uneven distribution of SNPs among 10 Kbp windows and the resulting effect on statistic distributions (Figure S5), we calculated significance thresholds for each statistic by binning windows with similar numbers of SNPs akin to methods by Voight et al. (2006). Bins of 10 Kbp windows were further subdivided into two groups based on the number of samples per population designation: (1) all pairwise comparisons of original population designations and (2) the two combined population comparisons (i.e., PRE–POST and NY–PA). All windows with 1–25 SNPs were disregarded for outlier identification due to low SNP density. For conservative determination of outlier regions (i.e., statistically significant 10 Kbp window), we calculated significance thresholds separately for each bin using a cutoff of 5 standard deviations above the mean bin value (see Tables S3.1 and S3.2 for binning designations). We reduced the confounding effect of localized geographic genetic differentiation by disregarding any preWNS–postWNS comparison outlier regions that overlapped with strictly geographic comparison (i.e., NY–PA, NYPRE–PAPRE and NYPOST–PAPOST) outlier regions with an additional 20 Kbp (i.e., within a 2 genomic window distance) buffer. Finally, we identified genes containing both F _ST and XP‐CLR outlier regions utilizing the mMyoLuc1 annotation and designated them as being under putative selection.

2.4.2. Soft Sweep Selection Discovery

We employed XP‐CLR and Rsb (Gautier et al. 2017) to identify signatures of soft selective sweeps. As XP‐CLR can identify both hard and soft selective sweeps (Vatsiou et al. 2016), the statistic was calculated and outlier regions identified as reported above. Rsb, the ratio of site‐specific extended haplotype homozygosity between populations, also takes the classic approach of assuming neutrality among the vast majority of loci, but measures the degree of departure based on the extent of linkage disequilibrium around a given site. To calculate Rsb, we first polarized the full SNP dataset and phased genotypes. SNP alleles were polarized using a combination of bcftools, vcftools and vcffilterjdk (Lindenbaum and Redon 2018), orienting variants with respect to the major allele of the PRE population. Next, genotypes were phased using BEAGLE v5.4 (Browning et al. 2021) on a per‐scaffold basis using the default recombination rate and a window size of 0.5 cM with 5% overlap for 40 iterations of 1000 phase‐states. Rsb was then calculated in 10 Kbp sliding windows with 10% overlap using rehh v3.2.2 (Gautier et al. 2017). We identified Rsb outlier regions using congruent methods as F _ST and XP‐CLR save for one detail—the significance threshold was set at 2.5 standard deviations above the mean bin value (see Table S3.3 for binning designations) due to the 2‐sided distribution of the statistic (Figure S5C). Identical to hard sweep detection, we omitted outlier regions that overlapped geographic outliers and identified genes under putative selection as those containing both XP‐CLR and Rsb outlier regions.

2.4.3. GO Term Enrichment

Genes identified by hard and/or soft sweep discovery methods were collated into a single list and underwent gene ontology (GO) enrichment analyses using the cytoscape v3.10.2 (Shannon et al. 2003) plugin cluego v2.5.10 (Bindea et al. 2009). We performed enrichment analyses using three separate ontologies: Biological Process (GO_BiologicalProcess‐EBI‐UniProt‐GOA‐ACAP‐ARAP_25.05.2022), Cellular Component (GO_CellularComponent‐EBI‐UniProt‐GOA‐ACAP‐ARAP_25.05.2022) and Molecular Function (GO_MolecularFunction‐EBI‐UniProt‐GOA‐ACAP‐ARAP_25.05.2022). For each ontology, we performed a right‐sided test to identify significantly enriched terms using Bonferroni step‐down correction to calculate p values. Significantly enriched terms (P ≤ 0.05) within each ontology were organized into a functionally grouped network using cluego and cluepedia v1.5.10 (Bindea et al. 2013). In these networks, nodes represent individual GO terms, edges link terms using kappa scores calculated based on the percent shared genes, and functional groups are represented by node colour. The final set of genes with strong evidence of selection for WNS survival were considered those associated with significantly enriched GO terms. We then performed PCA specifically on outlier regions that fell within this final set of genes to assess population structure preceding and following the arrival of WNS on those variants with the strongest and most cohesive evidence of selection.

2.4.4. A Priori Gene Network Analysis

We assessed functional and expressional interconnectedness among genes identified by this study as well as those discovered by previously published work exploring genetic responses to WNS by performing a gene network analysis using cluepedia. For this analysis we included our final set of genes as well as a list of 257 genes previously identified via literature review as being associated with WNS response through genome scan and/or gene expression analyses (Auteri and Knowles 2020; Davy et al. 2020; Donaldson et al. 2017; Field et al. 2015, 2018; Gignoux‐Wolfsohn et al. 2021; Lilley et al. 2017, 2019, 2020; Rapin et al. 2014), hereafter referred to as ‘a priori genes of interest’ (see Table S2). We annotated gene interaction edges based on the STRING v9.0 database (Szklarczyk et al. 2011) including metabolic (i.e., ‘reaction’, ‘binding’ and ‘catalysis’) and expressional (i.e., ‘expression’, ‘activation’, ‘inhibition’ and ‘post‐translational modification’) interaction types with a 0.3 minimum threshold value.

3. Results

3.1. Genomic Sampling and Population Structure

Whole‐genome resequencing of 59 individuals generated 21.9 billion cleaned reads, resulting in a mean depth of coverage of 14.0× –64.2× per individual (Figure S6) after alignment to the new chromosome‐level Myotis lucifugus reference genome (mMyoLuc1.0; GenBank accession: GCA_048340685.1). After extensive filtering, variant calling yielded 41.9 million high‐quality SNPs across 21 autosomes (genus Myotis 2n = 44, Bickham 1979) (Figure S7). When thinned to a minimum distance of 5 Kbp, the number remaining for analyses sensitive to linkage disequilibrium was 346,939 SNPs. Quality confirmation of the final SNP set showed no influence of sequencing library, mean depth of coverage, or percent missingness on genotype calls (Figures S8A, S8B and S8D). While differences in depth of coverage influenced genotype quality (GQ; Figure S8C), only 4.5% of SC genotypes fell below a value of 20 (the Broad Institute recommended minimum value for questionable datasets; Broad Institute 2024a), indicating no clear need for further filtering. Moreover, the range of variance in heterozygous SNP allele frequencies showed no outlier individuals (Figure S8E).

Estimates of genome‐wide population genetic diversity (π, values ranged 0.286–0.288) were similar across geographic origins and gave no indication of genome‐wide loss of variation following the arrival of WNS, similar to previous studies (Gignoux‐Wolfsohn et al. 2021; Lilley et al. 2020). Further, we found low genome‐wide pairwise population differentiation both when comparing across disease timepoints (F _ST: NYPRE–NYPOST = 0.0175, PAPRE–PAPOST = 0.0168, PRE–POST = 0.00843) and geographic origin (NY–PA = 0.00853; Table S4). Moreover, PCA indicated no clear pattern of population structure across space or time (Figure 1B). These results align with previous assessments of population structure in eastern populations of M. lucifugus where little to no structure has been detected (Gignoux‐Wolfsohn et al. 2021; Lilley et al. 2020; Wilder et al. 2015).

3.2. Detecting Signatures of Selection

While hard and soft sweep discovery methods were performed separately, consistencies emerged across all three statistics utilized (F _ST, XP‐CLR and Rsb). Within each statistic, significance cutoff values differed greatly both across SNP density bins and based on the number of samples per population comparison (Tables S3.1–S3.3). Additionally, all three statistics contained confounding geographic outliers (i.e., statistically significant windows identified in any NY–PA comparison) that fell within 20 Kbp of disease timepoint comparison outlier regions (Table S5). Finally, none of the statistics indicated that candidate sweep‐associated regions were genomically localized; outlier regions were spread across nearly all scaffolds. However, the number of outlier regions detected varied greatly by selection statistic and pairwise population comparison (Table S6), with Rsb identifying the most unique outlier regions (n = 10,252) and being the most susceptible to confounding geographic outliers (33.4% of windows), F _ST identifying the fewest unique outlier regions (n = 2267; 28.8% confounding outliers) and XP‐CLR being the least susceptible to confounding geographic outliers (27.2% of preWNS–postWNS windows; total of 4215 unique windows).

3.2.1. Hard Sweep Selection

Both statistics for hard sweep discovery (i.e., F _ST and XP‐CLR) identified vast numbers of regions under putative selection. F _ST indicated a total of 2267 preWNS–postWNS unique outlier regions across all but one scaffold (SUPER__21; Figure S9.1). After omitting those with confounding geographic comparison regions, 1613 unique outliers remained falling among 485 annotated genes. XP‐CLR detected a total of 4215 preWNS–postWNS outlier regions spread across all scaffolds (Figure S9.2), of which 3066 regions persisted once confounding geographic outliers were removed which identified 1031 genes. Comparison of the two sets of genes resulted in 170 target genes signifying the strongest evidence of hard sweep selection (Table S6).

3.2.2. Soft Sweep Selection

Statistics for soft sweep discovery (i.e., XP‐CLR and Rsb) similarly indicated many regions under putative selection. In this analysis, we considered the same outlier regions and genes identified by XP‐CLR. Rsb detected a total of 10,252 unique regions among all preWNS–postWNS comparisons found across all scaffolds (Figure S9.3). After omitting those confounded by geographic outliers, 6824 unique outlier regions remained identifying 1331 genes. Comparison of genes identified by both XP‐CLR and Rsb outlier regions resulted in 299 target genes, 84 of which were also detected by hard sweep methods (Table S6).

3.2.3. GO Enrichment

Across both hard and soft sweep detection methods, we identified 382 genes containing 4,P446 unique outlier regions with robust evidence of selection. GO enrichment analyses characterized broad‐scale gene function using three ontologies including Biological Process, which identified 58 enriched terms sorted into 10 functional groups (Figure 3A); Cellular Component, which identified 31 enriched terms sorted into 9 functional groups (Figure S10A); and Molecular Function, which identified 18 enriched terms sorted into 7 functional groups (Figure S10B). Across all ontologies, 201 genes (52.6%) were associated with enriched terms (Table S6), with a clear trend of enrichment in neuron development, organization and function (79.9% of terms; Tables S7.1–S7.3). SNPs within outlier regions exhibiting the strongest evidence of selection (n = 209,532) showed clear population structure across disease timepoints (Figure 3B), indicating considerable shifts in genomic variation following the arrival of WNS.

FIGURE 3.

Gene ontology enrichment and principal component analysis of outlier regions. (A) Shown is the network for significantly enriched (P ≤ 0.05, Bonferroni step down corrected) gene ontology (GO) terms utilizing the Biological Process database (see Figure S10 for other database results). Nodes represent individual terms and edges indicate shared genes among terms. Node colour indicates the broader functional group to which each term belongs, defined by high interconnectedness, while node size relates to the level of term significance (largest indicate P < 0.001). Large, coloured text indicates the most descriptive term of each functional group while smaller, grey text indicates all other associated terms. The pie chart in the upper righthand corner illustrates the percentage of terms belonging to each functional group. (B) Principal component analysis of gene variants under putative hard and/or soft selection associated with significantly enriched GO terms (n = 209,532). Pre‐ and post‐white‐nose syndrome populations (i.e., PRE and POST) are lumped across states. Ellipses are 95% confidence intervals and the percent variation explained by each principal component is shown in parentheses.

3.2.4. Interactions With A Priori Genes of Interest

Combining our final list of 201 genes with evidence of selection with 257 a priori genes of interest, eight of which directly overlapped (CAPN3, CEP112, GABRB1, NELL1, OLFM4, PRKCQ, SORCS3 and TRPM2), identified 15 distinct gene interaction networks (Figure 4A). Among these networks, 107 genes identified by this study (see Table S6) and 145 a priori genes of interest, representing all 10 studies (see Table S2), exhibited evidence of at least one gene interaction. Of the 15 networks that emerged from our analysis, one contained the majority of genes (85.1%). This large network was highly interconnected and displayed both metabolic and expressional interaction types (Figure S11).

FIGURE 4.

Gene interaction network of previously unlinked WNS adaptation study results. Each node represents an individual gene with the gene name indicated in text below. Edges represent direct gene interactions with the interaction type (i.e., gene expression or metabolic) not indicated (see Figure S11 for interaction types). (A) Node shape and colour depict the specific study of origin. Also shown are genes associated with (B) neural function, development and organization; (C) immune function; and (D) torpor and/or hibernation. For these representations of the gene network (B–D), colour represents the given biological association and shape indicates study of origin more broadly (circle = this study, square = a prior study, triangle = multiple studies).

4. Discussion

4.1. The Highly Polygenic Nature of WNS Adaptation

Although previous studies have attempted to capture the genomic effects of WNS on M. lucifugus in the eastern United States, we are the first to employ high‐coverage WGR alongside a chromosome‐level reference assembly and quantitatively incorporate findings from previous studies to evaluate the biological mechanisms and associated gene complexes underlying adaptive responses to WNS. Utilizing a rigorous suite of selective sweep detection methods, we identified a total of 382 genes with strong evidence of positive selection (Table S6). SNPs within these genes showed clear population structure across disease timepoints (Figure 3B), indicating changes in genomic variation following the arrival of WNS. These results indicate the genomic response to WNS in eastern M. lucifugus populations is highly polygenic. This is in stark contrast to all previously published genome scan studies, which identified very few and/or highly localized outlier regions (Auteri and Knowles 2020; Donaldson et al. 2017; Gignoux‐Wolfsohn et al. 2021; Lilley et al. 2020).

Of those genes associated with significantly overrepresented function (201), GO enrichment analyses illustrate a cohesive pattern. The majority of enriched terms (72; 67.3%) and their associated genes (191; 94.6%) were linked to neuron development, organization and function (Figure 3A and Tables S7.1–S7.3, also see Figure 4B). Moreover, when combining our findings with a priori genes of interest from 10 previous studies, we discovered multiple gene networks with the vast majority of genes (85.1% of 248 genes) belonging to a single, highly interconnected network (Figure 4A). This approach elucidated many direct interactions among genes of interest from all studies considered, including our own, and specifically revealed previously unidentified connections among prior studies. The number of genes and interactions identified by our gene network approach reinforces the polygenic nature of selection driving WNS survival in these populations.

Interpretation of the largest gene network further elucidates adaptations to WNS. This network signifies that a large component of adaptation is acting through immune response (Figure 4C). At the center where the density of interactions is highest are several cytokine‐coding genes (e.g., IL1B, IL6, IL10 and TNF), which act as important secondary messengers in inflammation and immune response signalling (Brocker et al. 2010). Additionally, there are several toll‐like receptor (TLR) genes (e.g., TLR2, TLR4 and TLR9) known to have key functions in immune signalling and infection defence (Vijay 2018). Among all gene networks identified we see further support for the importance of neuron function and development (Figure 4B). Approximately half (105; 55.0%) of the genes associated with enriched neuron terms (see Tables S7.1–S7.3) are spread across the largest network and comprise most of the remaining, smaller networks. Finally, we find further connections between torpor regulation and adaptive responses to WNS. Throughout the main network, 69 genes, including those from both our and previous studies' results, have shown significant associations with torpor regulation (Figure 4D; see Supplementary Methods S2 for more details) by multiple gene expression studies (Coussement et al. 2023; Cussonneau et al. 2021; Fedorov et al. 2009, 2011, 2014; Gillen et al. 2021; Gracey 2004; Mugahid et al. 2019; Weir et al. 2024; Xu et al. 2013; Yang et al. 2023).

Altogether, patterns of gene function and the degree of interconnectedness indicated by gene network analysis strengthen not only our own findings but also those of previous WNS response studies. Although very little direct overlap exists among the actual genes of interest identified by studies of WNS survival, we see that outcomes are not necessarily conflicting. On the contrary, they appear to be individual pieces of a larger puzzle of how selection has acted on a highly complex trait. Our findings illustrate that combining the efforts of multiple genomic sequencing techniques and gene expression analyses reveals the multifaceted mediation of adaptation to WNS in eastern M. lucifugus populations.

4.2. Possible Mechanisms of Adaptation to WNS

While neurons are involved in a vast number of biological functions, there are several clear pathways through which they may mediate adaptation to WNS. Many studies have found strong links between neuron function and skin wound healing, a clear candidate for conferring WNS survival given the pathogen, Pd, primarily infects epidermal and dermal tissues (Meteyer et al. 2009). Sensory neurons, specifically nociceptors (i.e., pain sensory neurons), have been shown to significantly impact injury‐induced inflammation and skin wound healing (Lu et al. 2024), help maintain skin barrier function by initiating neuropeptide and chemokine signalling (Hanč et al. 2023), and even mediate host defence against skin‐infecting fungal pathogens via dendritic cell activation and cytokine expression, specifically interleukin 23 (IL‐23; Kashem et al. 2015).

Neuron function has also been associated with torpor regulation, a major biological function linked to WNS pathology as arousal frequency is indicative of mortality rates (Lilley et al. 2016; Reeder et al. 2012). Glutamatergic neuron activity, a specific function enriched in our analyses (see Tables S7.1–S7.3), is involved in torpor initiation and arousal control (Hrvatin et al. 2020). In addition, sensory neuron thermosensitive transient receptor potential channels are differentially expressed during torpor in Myotis species (Li et al. 2022). Finally, glutamatergic and GABAergic neurons, another specific neuron type identified by our enrichment analysis (see Tables S7.2 and S7.3), have been associated with brown adipose tissue thermogenesis regulation (Morrison et al. 2014), which is critical for arousal from torpor (Neuweiler 2000). We postulate that connections between neuron function, skin wound response and torpor regulation suggest a complex adaptive mechanism in which perception of epidermal damage is altered in a way that simultaneously reduces cupping erosion damage (i.e., increases wound healing) while maintaining torpor (i.e., reduces arousal frequency). Additional physiological and transcriptional studies are needed to verify this theory.

Given the complexity of WNS morbidity in M. lucifugus , it follows that adaptive responses would develop through several mechanisms. Consequentially, our results emphasize the importance of employing methods, like those illustrated here, which take a comprehensive approach to identifying genes with evidence of selection for complex, polygenic traits. It is important to recognize that we have characterized the broad, rapid genomic response to WNS in the early stages of demographic recovery. Closely related Palearctic Myotis species, which have coexisted with Pd for at least a century (Campana et al. 2017), show a disparate response to Pd exposure, exhibiting mild symptoms of Pd infection (Kovacova et al. 2018; Zukal et al. 2016) despite comparable Pd loads (Hoyt et al. 2020; Zukal et al. 2016). Combined with a lack of transcriptional response to infection (Lilley et al. 2019), it appears that tolerance has emerged through long‐term coevolutionary mechanisms (Fischer et al. 2020). As populations of M. lucifugus enter the later stages of epidemic and Pd establishes as an endemic pathogen, it is possible that the dynamics of resistance and/or tolerance that have afforded immediate adaptive responses to Pd may shift for long‐term population survival (Harazim et al. 2026; Roy and Kirchner 2000; Whiting‐Fawcett et al. 2025). Further investigations will be needed to assess how evolution unfolds in M. lucifugus populations persisting with long‐term Pd exposure.

While Pd continues to spread across North America (Hoyt et al. 2021), it is inevitable that virtually all populations of M. lucifugus , and Nearctic Myotis generally, will experience some degree of pathogen burden. Given the variability in local ecosystems, hibernation strategies and intrinsic susceptibility to Pd across North American bat species (Frick et al. 2017; Hoyt et al. 2021; Moore et al. 2018), we emphasize that our results specifically relate to the genomic response in Eastern M. lucifugus . However, our results may pave the way for future studies to delve deeper into the biological pathways associated with adaptation to WNS in susceptible Nearctic species and subspecies across the expanse of Pd spread throughout North America.

4.3. The Importance of Whole Genome Resequencing and Soft Sweep Detection

The difference in the number of genes identified between ours and other studies is undoubtedly due to the sequencing method applied. Some studies (Auteri and Knowles 2020; Donaldson et al. 2017) used targeted or reduced representation sequencing methodologies, which for some systems may not have the resolution to adequately detect signatures of selection (Lowry et al. 2017). The remaining studies (Gignoux‐Wolfsohn et al. 2021; Lilley et al. 2020) performed variations of whole genome sequencing. However, one lacked individual‐level data (Lilley et al. 2020), which can alter allele frequency estimates and limits the types of selection tests that can be employed (Anderson et al. 2014). The final study (Gignoux‐Wolfsohn et al. 2021) performed low‐coverage WGR (~2.41× per‐individual mean versus 25.81× for this study), which generates genotype likelihoods rather than hard calls. This form of genomic data can lead to spurious genotype calls and limits analytical approaches (Lou et al. 2021). Given the ever‐decreasing costs for genomic sequencing, we recommend future studies employ high‐coverage (≥ 20×) WGR for reliable variant and genotype hard calls (Lin et al. 2024), which in turn offers the broadest range of analytical approaches (Bourgeois and Warren 2021).

Robust studies of genomic adaptation to novel pathogens also demand chromosome‐level assemblies to optimize interpretation of sequencing data. All previous studies were dependent on the previous M. lucifugus reference genome published in 2010 (Myoluc2.0, GenkBank accession: GCA_000147115.1). This assembly is highly fragmented, comprised of 11,654 scaffolds with an N50 of 4.3 Mbp (versus 258 scaffolds with an N50 of 99 Mbp for the new assembly; Figure S1), which reduces mapping quality for population‐level data (Stephens and Iyer 2018), and thus compromises data accuracy. Further, the previous assembly has 6034 fewer annotated genes (24,859 genes versus 30,893 for the new assembly) and many (9662) differing in size by at least 1 Kbp. These differences in gene annotation limit interpretation of biological consequences of changes in genomic variation. Our new chromosome‐level reference assembly (mMyoLuc1) in conjunction with high‐coverage WGR offered extremely high confidence in SNP and genotype calls, expanding the ability to perform more accurate and comprehensive analyses of selection.

Utilizing optimal genomic resources to attain accurate results is especially important in systems where results can have direct implications for on‐the‐ground management practices. WNS has continued to decimate populations throughout much of North America as it spreads westward to naïve populations along the Pacific coast (Cheng et al. 2021; Hoyt et al. 2021), demanding further intervention and mitigation efforts. Highly complex, quantitative traits are far more likely to be subject to complex, polygenic selection (Barton et al. 2017). This has specifically been found to be true for the development of disease resistance (Poland and Rutkoski 2016). Spurious or oversimplified results classifying mechanisms of adaptation could lead to wasted resources on limited laboratory methodologies and/or misguided therapeutics. By employing cutting‐edge genomic resources and comprehensive analyses in detecting adaptive genetic variants, we optimize the likelihood of achieving accurate results and, ultimately, improving the efficacy of resultant management strategies.

A considerable number of genes (84; 21.8%) in our study were detected by both hard and soft sweep methodologies. This may be explained by the fact that signatures of hard and soft sweeps are not always distinct (Vatsiou et al. 2016), as both result in strong patterns of linkage disequilibrium surrounding the site under selection (Pennings and Hermisson 2006). Further, a major distinguishing factor of these two modes of selection is the degree of polymorphism surrounding the site under selection—soft sweeps tend to maintain polymorphisms while hard sweeps eliminate them (Hermisson and Pennings 2005; Pennings and Hermisson 2006). However, if there are few polymorphisms surrounding the focal site prior to the selective pressure, soft sweeps may be less distinguishable from hard sweeps.

Given the brief amount of time (~10 years, or ~5 generations) for selection to act in the given sampling period of this study, we expected that selection was far more likely to act on standing variation than a single new beneficial mutation. This hypothesis was supported by the fact that most genes under putative selection (77.7%) were detected by soft sweep methodology. For adaptive selection in response to WNS to develop solely through means of hard sweep selection, new beneficial mutations would need to arise almost immediately and escape the stochastic effects of drift, which acts more strongly on small populations (Crow and Morton 1955). Standing variation segregated among multiple individuals is more likely to survive a population bottleneck and contribute to adaptation to new conditions far more quickly than a single new mutation (Hermisson and Pennings 2005), theoretically making this the more common mode of selection in natural populations (Messer and Petrov 2013). While few comparative studies exist, those that have been performed in both experimental and natural populations have found soft sweeps to be more common than hard sweeps (e.g., Garud et al. 2015; Hejase et al. 2020).

4.4. Genome‐Wide Patterns of Population Differentiation

The overall lack of geographic population structure suggests that individuals from NY and PA represent a single, interbreeding population (Figure 1B & Table S4). These results were perhaps unsurprising given a general absence of genetic structure has previously been reported for eastern populations of M. lucifugus (Gignoux‐Wolfsohn et al. 2021; Lilley et al. 2020; Wilder et al. 2015). This lack of population structure is likely a reflection of high migration rates and connectivity via mass hibernacula congregations (Frick et al. 2010; Norquay et al. 2013), which in turn likely explains the rapid spread of WNS in the eastern United States (Hoyt et al. 2021). While high connectivity can theoretically undermine adaptive shifts in the genome (Wright 1931), strong selective pressure, as clearly observed in WNS, can overcome the homogenizing effects of gene flow (Tigano and Friesen 2016). Moreover, dispersal and migration may even facilitate broad‐scale adaptation by spreading advantageous alleles across demes (Morjan and Rieseberg 2004), diluting the effects of drift. This is especially true for species with high N _e to N _c ratios (Delord et al. 2024), like Myotis bats (Wright et al. 2021).

Despite extensive records of M. lucifugus population loss following the arrival of WNS (Hoyt et al. 2021), we detected no significant loss of genome‐wide diversity in post‐WNS individuals (Table S4), reflecting patterns found in other whole‐genome studies (Gignoux‐Wolfsohn et al. 2021; Lilley et al. 2020). One possible explanation could be insufficient time for drift to have eliminated a significant amount of genome‐wide diversity. Although rare alleles are lost immediately following a population bottleneck, loss of heterozygosity via drift while the population remains small can take generations to occur (Maruyama and Fuerst 1985). Moreover, the detectable effects of drift can be delayed in long‐lived species (Gargiulo et al. 2024) such as M. lucifugus (Brunet‐Rossinni and Austad 2024). Despite their longevity, M. lucifugus breed every year (Frick et al. 2010), leading to overlapping generations which obscures how drift will act on populations over time (Ellner and Hairston 1994). While uncommon, examples of known demographic contractions not being reflected in genetic data have been documented in wild populations (Assis et al. 2013; Busch et al. 2007; Pujolar et al. 2011). Conversely, local loss of diversity may have been masked by surviving individuals coalescing into fewer hibernacula following widespread population loss. While site fidelity has historically been high for M. lucifugus (> 90%; Lewis 1995), long‐distance dispersal has been documented (Norquay et al. 2013) and WNS has been shown to impact community structure (Frick et al. 2015). Further, congregation at fewer sites has been supported by anecdotal observations from those managing eastern populations (Scafini and Turner 2018).

4.5. Non‐Model Selection Studies: Call to a Paradigm Shift

Harnessing the power of pairwise population differentiation statistics has become a widely popular method for detecting and assessing signatures of selection in wild populations. As such, trends in how these studies are conducted in non‐model species have emerged (Haasl and Payseur 2016) that now appear relatively fixed in relation to the progression of analytical techniques (Pavlidis and Alachiotis 2017). Incorporating these novel tools on an array of species and systems would establish more diverse methodologies of assessing signatures of selection in contemporary populations and, ultimately, expand our knowledge regarding processes of natural selection. Here, we outline several avenues we have employed that would improve the scope of future selection studies.

Classical implementations of selective sweep analyses, like pairwise F _ST, are effective in detecting signatures of hard sweep selection (Manel et al. 2016) and are therefore some of the most popular methods utilized. However, soft sweeps are likely the more common mode of selection, especially for recent phenomena (Hermisson and Pennings 2005, 2017; Messer and Petrov 2013). As a number of statistics have emerged that successfully identify signatures of soft sweeps, such as rehh (Gautier et al. 2017) and selcan (Szpiech 2024), such approaches must become standard practice. This is not to say that soft sweep analyses should supplant conventional methods—hard sweep detection has been invaluable for many systems (Jensen 2014). Consequently, many studies would benefit from utilizing hard and soft sweep approaches concurrently, as we did here. Of course, all selection statistics, regardless of the focal mode of selection, possess limitations and therefore benefit from being paired with one or more statistics to compensate for shortcomings (Vatsiou et al. 2016) and increase outlier identification stringency to minimize false positives.

Moreover, selection studies would benefit from shifting focus from few genes of large effect to methods geared towards detection and interpretation of polygenic selection. Traditional approaches often aim to identify genes with the largest signals and/or most interesting function in relation to the trait of interest. However, decades of genomic sequencing research has revealed that complex traits are overwhelmingly controlled by genetic variation in many loci of small effect (Barghi et al. 2020). By incorporating multiple statistics able to detect more than one type of selection signature, we can identify many loci of small effect without excluding the ability to detect genetic variants of large effect. Further, it is critical that genes of interest identified by previous studies within the given system are quantitatively incorporated. Often, results are compared in a narrow, qualitative fashion, with connections being drawn only when the same individual genes are identified. We have demonstrated that a lack of direct overlap in genes of interest among studies does not necessarily translate to conflicting results. While interpretation of the biological mechanisms of selection may seem daunting when faced with potentially hundreds of genes of interest, quantitative analyses of shared function and gene interactions offer the ability to delineate cohesive biological pathways of adaptation. Additional analytical approaches beyond the scope of this study may further illuminate mechanisms of adaptive evolution, such as landscape genomics approaches like environmental association analysis (Rellstab et al. 2015), SNP annotation to identify amino acid sequence changes and/or variants in gene regulatory elements (e.g., Cingolani et al. 2012), or investigations into changes in insertions/deletions and/or copy number variants (e.g., Li et al. 2023).

4.6. Genomics and Conservation Management

Our study serves as a strong example of the possible applications of population genomics in conservation management. We hope our work will facilitate improving current methods or illuminate novel avenues of WNS mitigation, intervention and therapeutics. The differences in genetic variation (Vonhof et al. 2015; Wilder et al. 2015), species composition (Patten 2004; U.S. Geological Survey 2016) and hibernation behaviour (Weller et al. 2018) in western Myotis species make predicting the effects of WNS difficult. Our results may, however, be an indication of the high adaptive potential of Myotis species, providing hope that naïve populations may have the capacity to respond quickly to WNS through rapid adaptation.

While the utility of genomics in wildlife management has been proven both in identifying imperilled populations (McCartney‐Melstad et al. 2018; Peters et al. 2016; Ruegg et al. 2018) and assessing the efficacy of in situ management practices (Capel et al. 2022; Grossen et al. 2018; Weeks et al. 2017), implementation within management agencies is still limited. Given that agency capabilities vary, integrating genomics could be achieved by simply incorporating currently published findings into action plans, collaborating with external genomics experts (e.g., university researchers), or even establishing their own research units (e.g., CDFW and USGS). As anthropogenically‐driven extinction events mount, the best chance we have to ensure management efficacy is to utilize every available tool to conserve our vulnerable wildlife.

Author Contributions

Conceptualization and design of this study was performed by Michael R. Buchalski, Thomas M. Lilley, Amy L. Russell, Maarten J. Vonhof, Samantha L.R. Capel and Devaughn L. Fraser. Samples for this study were provided by Thomas M. Lilley, Kenneth A. Field and DeeAnn M. Reeder. In‐house laboratory work was performed by Devaughn L. Fraser. Study methodology was performed by Samantha L.R. Capel with support from Devaughn L. Fraser and Juan Manuel Vazquez. Data analysis and visualization was performed by Samantha L.R. Capel. The original draft of this manuscript was written by Samantha L.R. Capel while reviewing and editing was performed by Samantha L.R. Capel, Devaughn L. Fraser, Kenneth A. Field, DeeAnn M. Reeder, Amy L. Russell, Peter H. Sudmant, Juan Manuel Vazquez, Maarten J. Vonhof, Thomas M. Lilley and Michael R. Buchalski.

Funding

United States Fish and Wildlife Service White‐nose Syndrome Recovery Research Grant Program award no. F19AP00284.

Ethics Statement

This study focused on bats from a non‐endangered species in strict accordance with recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All methods were approved by the Institutional Animal Care and Use Committee at Bucknell University (protocols DMR‐016 and DMR‐17) and performed under Pennsylvania Game Commission Special Use Permit #33085 and New York State Department of Environmental Conservation Permit #427.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Sample information.

Table S2: A priori genes of interest regarding white‐nose syndrome response.

Table S3.1: F _ST outlier window summary.

Table S3.2: Normalized XP‐CLR outlier window summary.

Table S3.3: Rsb outlier window summary.

Table S4: Population summary statistics.

Table S5: Total number of outlier regions.

Table S6: Genes under putative selection.

Table S7.1: Significantly enriched biological process gene ontology terms.

Table S7.2: Significantly enriched cellular component gene ontology terms.

Table S7.3: Significantly enriched molecular function gene ontology terms.

Figure S1: Reference genome scaffold size comparison.

Figure S2: Raw SNP quality metrics and filtering cutoffs.

Figure S3: Mappability visualized as mean read depth vs. mapping quality.

Figure S4: Sample kinship as calculated by SNPRelate.

Figure S5: Genome‐wide distributions of each selection statistic.

Figure S6: Per‐individual mean depth of coverage.

Figure S7: Autosomal SNP filtering.

Figure S8: Final variant quality metrics.

Figure S9.1: Genome‐wide F _ST for all pairwise comparisons.

Figure S9.2: Genome‐wide normalized XP‐CLR for all pairwise comparisons.

Figure S9.3: Genome‐wide Rsb for all pairwise comparisons.

Figure S10: Gene ontology enrichment networks.

Figure S11: Gene interaction network indicating interaction type.

Data Availability Statement

Raw data for this study are available through the NCBI Short Read Archive (BioProject PRJNA1353610, accession nos. SAMN52933055–SAMN52933113). All other data files and code are available through Dryad (https://doi.org/10.5061/dryad.ncjsxkt66) and GitHub (https://git.new/wl10DcJ), respectively. The genome annotation utilized for analyses is available through GitHub (https://github.com/docmanny/myotis‐gene‐annotations).