Vast alteration of the biosphere by humans is causing a sixth mass extinction. Adaptation to modified environments often is the only means for species to persist, but the extent to which such "evolutionary rescue" can prevent or reverse biodiversity loss is largely unknown. Using results from genomic analyses, a 15-year reintroduction effort, and population modeling, we provide a compelling example of evolution reversing the declines of an imperiled amphibian and allowing the reestablishment of extirpated populations. The once-common mountain yellow-legged (MYL) frog is threatened with extinction by the human-facilitated emergence of a lethal fungal pathogen (Batrachochytrium dendrobatidis; "Bd"). Although most MYL frog populations are extirpated following disease outbreaks, some persist and eventually begin to recover. Using an exome capture approach, we show that MYL frogs have undergone substantial evolutionary change following disease outbreaks. Moreover, in naturally recovering populations observed changes in immune function loci are associated with increased resistance/tolerance to Bd infection. Large-scale reintroduction of frogs from rescued populations resulted in the establishment of reproducing populations despite ongoing disease. In addition, results from viability modeling suggest that established populations have a low probability of extinction over 50 years. Collectively, these results provide a rare example of how evolutionary rescue and the reintroduction of resistant/tolerant individuals can allow the landscape-scale recovery of disease-impacted species. This example has potentially broad implications for the many taxa worldwide that are threatened with extinction by anthropogenic stressors.

Raw reads were filtered for adapters and contaminants using fastp and aligned to the Rana muscosa genome (NCBI SRA: SRS6533475), with repetitive elements masked using bwa (“mem” mode). Exact PCR duplicates were marked using Picard. Variants were then called following GATK best practices (v.4.2.0.0). Briefly, raw variants were called for each sample using HaplotypeCaller and combined using CombineGVCFs. Next, genotypes were jointly called using GenotypeGVCFs. Variants were then hard filtered using gatk VariantFiltration using the following parameters to remove low-quality sites: QD < 2.0, FS > 40.0, SOR > 3.0, MQ < 50.0, MQRankSum > 3.0, MQRankSum < -3.0, ReadPosRankSum > 3.0, ReadPosRankSum < -3.0. This initial filter resulted in 1,595,206 variant sites across 53 individuals. We then further filtered our dataset at the individual and variant level. First, we trimmed our variants to only include those with minor allele frequency > 0.03, a maximum depth of 250 and minimum depth of 5, a minimum genotype quality of 20, and a maximum missing proportion of 0.5. This filter resulted in 427,038 sites, of which 353,172 were SNPs and 73,866 were INDELS. Finally, we trimmed samples with an average depth across filtered sites < 7x. Our final dataset included 50 samples, 23 from naive and 27 from recovering populations, with an average depth of 16.7x (range = 7.4x – 26.1x).