Habitat restoration can help to protect and support endangered species. Following the restoration of Lake Constance to oligotrophic conditions, deep-water charr (originally described as Salvelinus profundus Schillinger, 1901) were rediscovered after more than 40 years of presumed extinction caused by eutrophication. Subsequent targeted surveys lead to the capture of a further 127 deep-water charr and 233 sympatric normal charr, Salvelinus cf. umbla. Given a history of intensive stocking in the lake and the possibility of introgressive hybridization, a discussion arose about the genetic status and future conservation strategies of both forms. The current study gathered life history traits and morphological and population genomic data for both extant forms, for comparison with 60–120 years' worth of historical data and museum specimens. The extant deep-water charr resembled historical specimens in terms of body shape, but significant differences between historical and extant normal charr included a decline in gill raker number over time and distinctly different body shapes and growth patterns. RAD-based population genomic analyses of the contemporary samples support these results, revealing the two extant forms to be highly divergent, albeit with a few putative hybrids. SNPs-based population assignment tests of historical and contemporary samples strongly suggest that deep-water charr persisted unaltered in Lake Constance during the eutrophic phase. Meanwhile, none of the historical genomic samples could be assigned to the extant normal S. cf. umbla population. Functional analyses of gill raker number and stocking intensity pointed to stocking with non-endemic charr as the most likely reason for the disappearance of the endemic normal form. These results show that interplay between different anthropogenic impacts, i.e. eutrophication and stocking can lead to irreversible but hard-to-detect impacts on endemic biodiversity. Understanding these impacts is a prerequisite for effective conservation of endangered populations of cryptic fish.

Dataset_1: RAD sequencing data. Salvelinus DNA-extractions were sent to Floragenex (Portland, Oregon, USA) for SbfI-RAD-Seq library preparation. One 95-plex library containing all recently collected charr individuals from Lake Constance was sequenced twice on two Illumina HiSeq (1x100bp) lanes. Returning fastq-files were demultiplexed on the basis of 10bp individual barcodes, which were removed after demultiplexing. They were quality filtered (phred score -33) using the process radtags 2.4 script of the stacks 2.4 program package (Catchen et al. 2013). All reads of the same individual of both lanes were merged using a custom script. The program cutadapt v.1.12 (Martin 2011) was used to trim the first 6 bp cutting sites of each read.  

Dataset_2: Based on the locus-specific F-statistics we initially selected 182 RAD locus sequences featuring comparatively high F_st-Values between deep-water charr and normal charr phenotypes and containing only a single “diagnostic” SNP. Those 182 loci were clustered at 86% similarity in order to remove candidate bait loci with too high similarity; further, these loci were checked with the same script for low complexity due to microsatellites or other short repeats. 164 candidate bait loci were retained and sent together with DNA extracts from historical scale samples to the Arbor Biosciences lab (Ann Arbor, Michigan, USA), which BLAST-filtered 99 out of 164 loci for bait design that would not target overrepresented regions in the charr genome. Based on this selection 497 baits were designed, each 60 bp in length and spaced every 15 bp along RAD loci and representing these 99 loci. Subsequent laboratory steps performed by Arbor BioSciences comprised hybridization of baits to the 99 target loci in the double-stranded DNA libraries constructed from DNA extracts, amplification of captured libraries and sequencing on an Illumina HiSeq2500 platform (paired end, 2x150 bp). Sequencing reads of library-prepped individuals were downloaded and corresponding reads of identical individuals merged using a custom script. After visual quality check in FASTQC 0.11.3 (Andrews 2010), we used cutadapt v.1.12 (Martin 2011) to remove the first 15 bp of all reads (including the 13bp adaptors), to filter out only reads with a minimum length of 25 bp long reads. We further trimmed all reads to a maximum length of 100bp.

Dataset_3: 99 SbfI-RAD derived target loci selected based on locus-specific F-statistics (for details, see MS)

Dataset 1: 89 demultiplexed, quality filtered and trimmed 85bp .fastq-files derived from SbfI_RAD libraries run on two Illumina HiSeq (1x100 bp) lanes. Individuals are labeled according to phenotypic assignment of samples to either "Tiefseesaibling"  (N=46) or "Normal" (N= 43) in combination with sample-ID numbers of the Fisheries Research Station Baden-Württemberg.

Dataset 2: 14 demultiplexed, quality filtered trimmed to 100bp .fastq-files mapped to the Salvelinus sp. GCF_002910315.2 reference genome derived from Illumina HiSeq2500 runs of hybridisation capture libraries targeting 99 RAD loci differentiating between "Tiefseesaibling" and "Normal"-Saibling, i.e. the extant Salvelinus forms in Lake Constance.

Dataset 3: 99 SbfI-RAD derived target loci selected based on locus-specific F-statistics.