Climate change is increasingly impacting temperate forest ecosystems and many forest herbs might be unable to track the changing climate due to dispersal limitation. Forest herbs with a low adaptive capacity may therefore benefit from conservation strategies that mitigate dispersal limitation and evolutionary constraints, such as assisted migration. However, assisted migration strategies rarely consider evolutionary constraints of potential source populations that may jeopardize their success. In cases where climate adaptation is overshadowed by competing evolutionary processes, assisted migration is unlikely to support adaptation to future climates. Using a combination of population and landscape genomic analyses, we disentangled local adaptation drivers and quantified the adaptability and vulnerability to climate change of the self-incompatible deciduous forest herb Primula elatior. Southern populations displayed a sharp genetic turnover, and a considerable amount of local adaptation under diversifying selection was discovered. However, most of the outlier loci could not be linked to climate variables (71%) and were likely related to other local adaptation drivers, such as photoperiodism. Furthermore, specific adaptations to climate extremes, such as drought stress, could not be detected. This is in line with the typical occurrence of forest herbs in buffered climatic conditions, which can be expected to reduce selection pressures imposed by climate. Finally, populations in the south of the distribution area had increased sensitivity to climate change due to a reduced adaptive capacity and a moderate genetic offset, while central European populations were sensitive due to a high genetic offset. We conclude that assisted migration from southern source populations could bear significant risk due to nonclimatic maladaptation and a low adaptive capacity. Regional admixture and restoration of ecological connectivity to increase the adaptive capacity, and assisted range expansion to suitable habitat in the north might be more appropriate mitigation strategies.

All libraries for each sequencing lane were demultiplexed using the Illumina `bcl2fastq v.1.8.4´ software (ILLUMINA, 2013). One mismatch or missing data point was allowed in the barcode read (5,343,298 ± 1,598,102 average total reads per sample ± SD). Sequencing adapter remnants were clipped from all raw reads, and reads with a final length < 20 bases were discarded (2,671,608 ± 799,040 average quality trimmed read pairs per sample ± SD). Adapter-clipped reads were quality trimmed by removing sequencing errors and trimming was focused on the 3'-end to get a minimum average Phred quality score of 10 over a window of ten bases (2,512,362 ± 747,790 average adapter clipped read pairs per sample ± SD). Merged quality trimmed reads were error corrected (201,600,000 total read pairs) using `Musket v.1.0.6´ with a 21 k-mer size for correction (Liu, Schröder, & Schmidt, 2013). Furthermore, digital normalization of error-corrected reads (148,503,886 total reads) was performed with the `normalize_by_median.py´ script from `khmer v.1.1´ with a k-mer size of 32 and a coverage cut-off of 80 (Crusoe et al., 2015).

De novo assembly was performed with the `CLC Genomics Workbench v.8.0´ (QIAGEN, 2021). Gene discovery was performed with `Augustus v.3.1´ (Stanke, Diekhans, Baertsch, & Haussler, 2008) on the post-processed scaffolds. Complementary draft functional annotation of predicted peptides was performed with `InterProScan v.5.4-47.0´ (Jones et al., 2014). InterPro lookups for pathway and GO annotation were performed based on the predicted peptides (Camon et al., 2005). Quality trimmed reads were aligned against Arabidopsis as reference genome using `BWA-MEM v.0.7.12´ (Burrows-Wheeler Aligner; Li, 2013; Li & Durbin, 2009). Variant discovery and genotyping of samples was executed with `Freebayes v.1.0.2-16´ with a diploid setting and reads with more than two mismatches were excluded from the dataset (Garrison & Marth, 2012). Annotations of variant effects on annotated genes and transcripts were performed using `SnpEff v.4.31´ (Cingolani et al., 2012). The predicted genes and transcripts from the genome annotation were used to predict downstream functional effects of the variants.

This SNP dataset of Primula elatior was filtered for a minimum read depth of 5 per SNP and a minimum allele frequency across all samples of 5%, but was not yet filtered for missing data. The data Format is the VCF data format. More info about this data format can be found at https://samtools.github.io/hts-specs/VCFv4.2.pdf.