Using genomics to guide seed‐sourcing at the right taxonomical level for ecological restoration projects: The complex case of Carex bigelowii s.lat. in Norway

Abstract There is a growing demand for ecological restoration using suitable seeds following international standards or national legal demands for local seed‐sourcing. However, before selecting the appropriate geographic origin of seeds, it is vital to explore taxonomic complexity related to the focal taxa. We used ddRAD‐seq to screen genomic diversity within Carex bigelowii s.lat. focussing on Norway. This species complex is considered a candidate for seeding, but presents considerable morphological, ecological, and genetic variation. The genetic structure of 132 individuals of C. bigelowii s.lat., including Carex nigra as an outgroup, was explored using ordinations, clustering analyses, and a genetic barrier algorithm. Two highly divergent clusters were evident, supporting the recognition of two taxonomic units “C. dacica” and C. bigelowii “subsp. bigelowii”. Previously defined seed‐sourcing regions for C. bigelowii s.lat. did not consider the known taxonomic complexity, and therefore interpreted the overall genetic structure as seed‐sourcing regions, not taxa. We estimated genetic neighborhood sizes within each taxon to be 100–150 km and 300 km, respectively, indicating species‐specific delimitations of local seed‐sourcing regions. Frequent hybrids, local genetic distinctiveness, and suggested ecotypes add complexity to the discussed seed‐sourcing regions. Our results show how genomic screening of diversity and structure in a species complex can alleviate the taxonomic impediment, inform practical questions, and legal requirements related to seed‐sourcing, and together with traditional taxonomic work provide necessary information for a sound management of biodiversity.


| INTRODUC TI ON
Species and ecosystems are threatened by widespread, continued, and accelerating degradation and loss, leading to loss of ecosystem functioning and services (IPBES, 2019). Therefore, the UN General Assembly has stated that "there has never been a more urgent need to restore damaged ecosystems than now," and declared the next decade 2021-2030 as the UN Decade on Ecosystem Restoration (https://www.decad eonre stora tion.org/). Rebuilding ecosystems can be done by ecological restoration, which is the process of assisting the recovery of an ecosystem that has been degraded, damaged, or destroyed (Clewell et al., 2004). Importantly, any restoration should consider all variability found among living organisms, including diversity within species, between species, and of ecosystems (CBD, 1992).
A growing demand for ecological restoration increases the requirements for seeds (Gann et al., 2019). The use of native, regional, or local seeds to prevent maladaptation and in-or outbreeding depression when reintroducing plants is considered most successful in ecological restoration (Breed et al., 2018;Hagen et al., 2014). However, there has been a shift away from traditional local seed-sourcing towards mixed and targeted approaches, broadening the genetic basis of a restored flora to maximize the long-term success and resilience of restoration programs (Bucharova et al., 2019;Hoffmann et al., 2021).
There are numerous proposed methods to evaluate the appropriate provenance and select a suitable seed-sourcing strategy (summarized in e.g. Breed et al., 2013Breed et al., , 2018. Recently, genomic approaches have been applied to identify hierarchically ordered clusters of genetic diversity and differentiation in plant species, providing useful tools for delimiting seed-sourcing regions (Breed et al., 2019;De Kort et al., 2014;Jørgensen et al., 2016;Williams et al., 2014).
An international standard for the use of native seeds in ecological restoration has just been published (Pedrini & Dixon, 2020), and seed transfer zones have been suggested for several countries (Bower et al., 2014;De Vitis et al., 2017), aiding restoration projects locally and nationally. In practical restoration of flora, there will be many biological unknowns when making decisions around seed-sourcing, but they will nevertheless have to be made within the prevailing legal framework. In Norway, ecological restoration projects must follow the legal framework set by the Norwegian Nature Diversity Act of 2009 (https://lovda ta.no/dokum ent/NL/lov/2009-06-19-100?q=natur mangf oldloven) with associated regulations (https:// lovda ta.no/dokum ent/SF/forsk rift/2015-06-19-716), which instruct the use of plant material of local provenance without defining what local means. Based on AFLP markers, general seed transfer zones for commercial seed production of a set of common and widely distributed alpine plant species in Norway have been proposed (Jørgensen et al., 2016). However, before selecting the appropriate geographic origin of seeds for re-introduction, it is vital to explore the taxonomic complexity related to the taxa in focus. By overlooking taxonomic complexity, one risks to misinterpret genetic structuring caused by the presence of taxonomic entities as seed-sourcing regions.
bigelowii), and plants growing in a dryer habitat (ridges) with a sessile staminate spike and more pointed pistillate scales (C. bigelowii subsp. dacica; Lid & Lid, 2005). Carex bigelowii subsp. bigelowii is described to be widely distributed in northern Norway and northeastern parts of southern Norway (Artsdatabanken, 2020b;Solstad et al., 2015; L. Galten, personal communication) while C. bigelowii subsp. dacica should be common throughout Norway. Thus, the current treatment of the two subspecies consider them morphologically and ecologically distinct, but broadly sympatric in Norway.
Unlike the described distribution of C. bigelowii s.lat. in Norway, three separate genetic studies including Scandinavian samples have all identified two distinct genetic clusters in the region; one in the north and one in the south. First, a northern Norwegian population of C. bigelowii subsp. bigelowii was found to fall into an American genetic cluster, and to be morphologically and genetically distinct from the European genetic cluster of C. bigelowii subsp. rigida (syn. subsp. dacica) containing a southern Norwegian population (Nakamatte & Lye, 2007). Likewise, a circumpolar phylogeographical study of C. bigelowii s.lat. found a genetic and geographical division of four Scandinavian populations into a northern group belonging to an amphi-Atlantic cluster (C. bigelowii subsp. bigelowii), a southern group belonging to a European cluster (C. bigelowii subsp. rigida), but also an admixed population in between (Schönswetter et al., 2008). Again, the same genetic pattern of a northern and southern genetic group within C. bigelowii [sic] was found by Jørgensen et al. (2016).
However, the latter two groups were not suggested to potentially represent the two subspecies, but instead hypothesized to represent different post-glacial immigration routes into Norway, and further suggested as two seed transfer zones for commercial seed production. However, analyses of carpological features among species in Carex section Phacocystis found unique morphological types of achenes for C. bigelowii s. str. and C. dacica, each with diagnostic characters (Jiménez-Mejías & Martinetto, 2013). Given the complex taxonomy of C. bigelowii s.lat., with taxa seemingly occurring in sympatry that are morphologically, ecologically, and genetically differentiated, we hypothesize that it cannot be treated as one taxon for delimiting intraspecific seed-sourcing regions in Norway.
The recent introduction of genomic methods to ecological restoration biology is considered a promising tool for achieving restoration goals, for example by identifying proper seed transfer zones and monitoring the genomic outcome of restoration (Breed et al., 2019;Williams et al., 2014). However, there are still very few case studies showing the potential of using genomic tools in guiding ecological restoration (but see Carvalho et al., 2021;Wood et al., 2020). Here, we use double digest RAD sequencing (ddRAD-seq) and genomewide SNP data to revisit the taxonomic uncertainty and discuss Norwegian seed-sourcing regions for C. bigelowii s.lat.by screening the genomic diversity of Norwegian populations of C. bigelowii s.lat.
with additional samples from North America and Europe. First, we investigate whether genetic structure found within C. bigelowii s.lat.
in Norway corresponds to the described subspecies, whether they have overlapping geographical ranges, and explore admixture within and hybridization between genetic groups. We further investigate genetic and possibly geographic structure and diversity within the two recognized subspecies in Norway, and discuss the impact of our results on delimitation of seed-sourcing regions.

| Sampling, library preparation, and sequencing
We collected 139 individuals of C. bigelowii s.lat. from 33 sites across its Norwegian distribution and 16 sites outside Norway ( Figure 1a, Table 1). In addition, three individuals of C. nigra (L.) Reichard were included as outgroup. Because C. bigelowii s.lat. is known to reproduce clonally (e.g., Callaghan, 1976), the individuals were sampled several meters apart, as far from each other as possible given the spatial extent of the populations. According to the differential morphological characters given in the Norwegian Flora (Lid & Lid, 2005), each Norwegian individual was tentatively determined to C. bigelowii subsp. bigelowii (eight sites), C. bigelowii subsp. dacica (20 sites), a hybrid between the two subspecies (three sites), or even a hybrid with C. nigra (two sites; Figure 2a, Table 1). All specimens collected outside of Norway were identified as C. bigelowii [sic] by the respective collectors.
Genomic DNA from silica-dried leaf material was extracted using NucleoSpin Plant II extraction kit (Macherey-Nagel) following the manufacturer's protocol, and quantified on a Qubit 2.0 using the HS Assay kit (Thermo Fisher Scientific), before ddRADseq libraries were prepared following the protocols described in Westergaard et al. (2019). One sample was replicated from DNA extraction through all analyses. The libraries were sequenced in four lanes of 81-bp paired-end reads on an Illumina NextSeq 500 at the Genomic Core Facility (GCF), Norwegian University of Technology and Science, Norway.

| De novo assembly of loci, variant calling, and filtering
The six part-libraries were demultiplexed and quality checked separately using the process_radtags component of STACKS v. 2.3e, before de novo assembly of a catalogue of loci, and calling of SNPs and genotypes with the component denovo_map (Catchen et al., 2011(Catchen et al., , 2013. The settings applied for the de novo assembly were chosen following the recommendations of parameter settings by Paris et al. (2017). In brief, our study is not phylogenetic, our data was clean (i.e., 98.68% perfect index and 97.50% retained reads), and samples were collected at different sites ("populations"); Altogether, 10 individuals collected from 10 different sites were excluded from the final analyses as they had more than 50% missing data (

| Genetic structure and diversity
The genetic structuring of the data was first explored for all individuals using dataset A in a principal coordinates analysis (PCoA) with Euclidean distance using the dartR package (Gruber et al., 2018) in RSTUDIO v. 1.0.44 (R Team, 2020). Next, we used clustering analyses to investigate the genetic separation of the morphologically defined taxa, that is, C. bigelowii subsp. bigelowii, C. bigelowii subsp. dacica, and the outgroup C. nigra, and potential indications of admixture between them (sensu Meirmans, 2015;van Hengstum et al., 2012). For this purpose, we first ran a Discriminant Analysis of Principal Components (DAPC) for all individuals using dataset A and the R package adegenet v.2.1.1 (Jombart, 2008;Jombart & Ahmed, 2011). Clusters were generated using the find.clusters() function (max no. 51 = all sites), 140 PCs were kept, and K = 3 was chosen based on the "elbow" of the Bayesian Information Criterion (BIC) curve. The dapc function was run on the three clusters identified, keeping 100 PCs and two discriminant analysis functions. Second, we ran STRUCTURE v.2.3.4 (Falush et al., 2007;Pritchard et al., 2000), using dataset B for K = 1-4 and including all individuals, with an a priori expectation of K = 3, while allowing for a potential fourth genetic cluster. We applied an admixture model, default settings with 10 replicate runs, each with 1,000,000 iterations and a burn-in of 100,000. STRUCTURE HARVESTER web v. 0.6.94 (Earl & vonHoldt, 2012) was used to summarize the runs. To implement geographical information in our assessment of overall genetic structuring of the data, the TESS3 algorithm  was applied for 10 runs per K (1-10) using dataset B and the TESS3r R-package (Caye et al., 2018). The "elbow" of the cross-validation score curve was evaluated to decide which K best describe the data, and K = 3 was visualized with a bar plot and interpolated on a geographic map. The dataset was scanned for outlier loci using an Fst approach implemented in the program Martins et al., 2016).
To assess how many SNPs support each interpreted taxonomic entity (i.e., taxon-specific SNPs), the "gl.filter.pa" function was applied with the dartR package using dataset A to calculate fixed (homozygote AA in one group and homozygote aa in the other group) and private alleles (heterozygote Aa in one group and homozygote aa) separating C. bigelowii "subsp. bigelowii" from "C. dacica," C. bigelowii "subsp. bigelowii" from C. nigra, and "C. dacica" from C. nigra. Sites with interpreted hybrids or uncertain taxonomic affiliation (7,11,15,21,22,29,33; see Table 1 "ID genomics," Figure 1) were grouped separately for these calculations.
To look for potential genetic barriers within C. bigelowii s. lat in Norway, the Monmonier maximum-difference algorithm (Monmonier, 1973) was used to detect boundaries within dataset B (excluding C. nigra) using the package adegenet in R. Two main genetic clusters were recognized within C. bigelowii s. lat. F I G U R E 1 Main genetic structure of sampled Carex bigelowii Torr. Ex Schwein s.lat. and the outgroup C. nigra (L.) Reichard in the North Atlantic region based on: (a) STRUCTURE results for K = 3, where "C. dacica" is interpreted as the red cluster, C. bigelowii "subsp. bigelowii" as the blue cluster, and C. nigra as the orange cluster; and (b) Twodimensional PCoA plot (first and second axes) based on dataset A (15,095 SNPs) for all 132 individuals and one replicate, where the geographical origin of each individual is shown with different shapes, and interpreted hybrid or taxonomically uncertain individuals are identified to sites (see Table 1 for site information)   To see if this site is better explained as a distinct cluster or an admixture of two or three clusters, we performed a test with sNMF allowing admixture, similar to what was performed using STRUCTURE by van Hengstum et al. (2012). To avoid bias occurring due to unbalanced sampling (Meirmans, 2019), the test was run with a selection of individuals from the three clusters recognized as our main taxonomic entities (four C. nigra, eight "C. dacica," nine C. bigelowii "subsp. bigelowii"), and the two individuals from site 33 Båtsfjord with unknown origin/taxonomy for K = 3 and K = 4, each with 10 replicates.
The "elbow" entropy values were considered to evaluate the genetic composition of site 33 Båtsfjord.

| Genetic clusters support two taxa of C. bigelowii s.lat. in mainland Norway
The PCoA of dataset A showed that individuals from southern Norway, Iceland, and Central Europe (hereafter referred to as "C. dacica") were highly separated from individuals in northern Norway, Canada, and Greenland (hereafter referred to as C. bigelowii "subsp. bigelowii") at the first axis (33.2%; Figure 1). The  Figures S12, S13). The calculations of taxon-specific SNPs found that most of the SNPs were specific for each interpreted taxon (i.e., C. bigelowii "subsp. bigelowii", "C. dacica," and C. nigra; Table S2), although there was some overlap. When K = 4, site 33 Båtsfjord in northern Norway formed a separate STRUCTURE cluster (data not shown). The two individuals were morphologically determined as C. bigelowii subsp. bigelowii, but seemed genetically admixed involving all three taxonomic entities when K = 3 (Figures 1, S8, and S10). Furthermore, this site formed a fourth cluster in a separate sNMF test ( Figure S14) and separated from all other individuals along the third axis of the PCoA ( Figure S15).
Individuals from the Icelandic site 51 Skipalon were morphologically identified as C. bigelowii [sic], but grouped closely with C. nigra in all analyses, and were thus considered misidentified. At site 29 Kvalsund, one individual was morphologically determined to C. bigelowii subsp. bigelowii, but genetically admixed between C. bigelowii "subsp. bigelowii" × C. nigra (Table 1;  were also observed ( Figure 1; Table 1). These individuals were not restricted to a geographic meeting zone in between the mainly northern C. bigelowii "subsp. bigelowii" and the southern "C. dacica." At the southern site 15 Leirtjønnkollen, six out of seven individuals were initially morphologically determined to be the northern C. bigelowii subsp. bigelowii, of which three formed a separate cluster, while three were interpreted as hybrids between "C. dacica" and C. bigelowii

| Genetic structure within C. bigelowii "subsp. bigelowii" and "C. dacica" in Norway
In northern Norway, the cross-entropy values for sNMF clusters plateaued between K = 2 and 3 ( Figure S17  interpreted as "C. dacica," three to a minor gene pool of their own, and three interpreted as the hybrid C. bigelowii "subsp. bigelowii" × "C. dacica" also found in two individuals from site 11 Forollhogna. Low levels of genetic structuring were found in southern Norway, and the fine-scale genetic analysis found that the southern individuals were spatially autocorrelated with an estimated genetic neighborhood size of ca. 100-150 km (Figure 2).

| D ISCUSS I ON
Knowledge about the speciation continuum is of particular importance when restoring degraded nature; otherwise, there is a risk of introducing foreign species and genotypes, biodiversity loss, and eventually, altered or even deteriorated ecosystem services (e.g., Alexander et al., 2016;Hagen et al., 2014). Our results illustrate how overlooked taxonomical complexity within a species complex, in this case one recommended for commercial seed production, affects the discussions and delimitation of seed-sourcing regions. Inaccurate species identification pertaining to ecological information is a fundamental impediment to reducing biodiversity loss and restoring functional ecosystems (CBD, 2008). Although species often are used as units for management, this is complicated for species complexes where taxa are difficult to distinguish and hybridization occurs.

| Overlooked taxonomical complexity is challenging
Our genomic screening focusing on Norway revealed that there are two unique and highly divergent evolutionary clusters within the common and widespread C. bigelowii s.lat., with a genomic distance even surpassing levels of species divergence found between C. bigelowii s.lat. and the outgroup C. nigra. Our results agree with the recent finding of two well-defined, distantly related phylogenetic groups recognized as C. bigelowii s.s. and C. dacica, clearly supporting the recognition of C. dacica as a separate species (Benítez-Benítez et al., 2021). Most authors consider C. bigelowii subsp. bigelowii and C. dacica to be closely related (e.g., Chater, 1980;Egorova, 1999;Nakamatte & Lye, 2007). However, the two clusters do not correspond to C. bigelowii subsp. bigelowii and subsp. dacica as described in the Norwegian Flora, supposedly occurring in broad sympatry (Lid & Lid, 2005). In contrast to reports of both subspecies being widely distributed, but ecologically differentiated (wet vs. dry habitats), we found them to be mainly geographically clustered in northern and southern groups. The northern group of C. bigelowii "subsp. bigelowii" clustered with individuals from Greenland and Canada, while the southern group of "C. dacica" clustered with individuals from Iceland and central Europe, following an overall pattern reported for European populations by Schönswetter et al. (2008) and Alsos et al. (2015).
Proper identification of Carex taxa often relies on few and subtle morphological characters, and demands a good overview of phenotypic plasticity within taxa and their ecology, often complicated by the frequent occurrence of partly or fully fertile hybrids (e.g., Cayouette & Catling, 1992;Nakamatte & Lye, 2010). Frequent hybridization within Carex section Phacocystis (including C. bigelowii s.lat. and C. nigra) further blurs taxon limits in this section (Duman & Kryszczuk, 1958;Lepage, 1956;Pedersen, 2017;Polunin, 1940;Raymond, 1952). Our results revealed a swarm of hybrids between C. bigelowii "subsp. bigelowii" and "C. dacica," and also with C. nigra, but hybrids were neither easily recognized morphologically (Lid & Lid, 2005; Table 1)  Carex bigelowii [sic] is reported to mainly reproduce by runners (Callaghan, 1976), but seedling recruitment has been observed in disturbed sites like new road verges and fresh river gravel bars both in mountains and in the Arctic (Jonsson et al., 1996;Schönswetter et al., 2008;Stenström, 1999). The main clonal reproduction may explain the reduced gene flow between the interfertile C. bigelowii "subsp.
bigelowii" and "C. dacica" in Norway, maintaining a general genetic structure with a southern and a northern group. Although the seeds of C. bigelowii s.lat. have no obvious morphological adaptations to long-distance dispersal, previous multilocus assignment tests have revealed postglacial, trans-Atlantic dispersal of both subspecies , indicating a long-distance dispersal potential.
The site 33 Båtsfjord is located in the northernmost part of Norway, far away from any known "C. dacica" population, but our results indicate that the individuals may be admixed of C. bigelowii "subsp. bigelowii," "C. dacica" and C. nigra. This site is genetically distant from near-by site 31 Sør-Varanger (Figure 2), and Jørgensen et al. (2016) also found another nearby population to be genetically "completely separated from the remaining." They hypothesized a polemochoric (i.e., war dispersed) origin, as many species were introduced to this region during World War II (Alm et al., 2009).
Few case studies report on the consequences of undeliberate seeding of maladapted cryptic taxa or genetic lineages during ecological restoration, which might reflect that unsuccessful seeding is not reported, or the cause of failure is not identified (Hobbs, 2009).
However, different negative effects of using commercially produced seed mixtures have been observed at restored sites (e.g., Aavik et al., 2012;Hagen et al., 2014), and several ecological studies report on intraspecific differences without reflecting on taxonomic uncertainties. For example, C. bigelowii [sic] is often considered a model species for long-lived, arctic-alpine clonal species because of the extensive data that exist on its population dynamics and demographic responses to a range of environmental factors (e.g., Carlsson & Callaghan, 1994;Jonsson et al., 1996;Little et al., 2015). However, by neither exploring nor acknowledging taxonomic complexity, studies on C. bigelowii [sic] including sites from different continents may have been conducted on different taxa reported by the same name, which in retrospect may cause confusion and even question the validity of interpreted results. For instance, Stenström (1999) found differences in the sexual reproductive ecology of C. bigelowii [sic] in northern Sweden and Iceland, and discussed how climatic or genetic differences affecting pollination and pollen viability might explain this, while Little et al. (2015) also treated C. bigelowii [sic] as a species, and reported highly species-specific responses to global changes. Our results fall in line with a previous study (Schönswetter et al., 2008), supporting the recognition of Icelandic populations as C. dacica, while northern Swedish populations are most probably C. bigelowii subsp. bigelowii, or potential hybrids.

| Implications for defining seed-sourcing regions
Species delimitation is a highly contested area, and there is disagreement about almost every aspect of how to define and use the "species" category (see e.g., Padial & De la Riva, 2021;Raposo et al., 2020;Stanton et al., 2019). A thorough taxonomic and nomenclatural revision of the entire C. bigelowii species complex is clearly needed, but our results can be used to delineate seed-sourcing regions relevant for ecological restoration in Norway under the assumption that genome-wide likeness is a useful criterion for this purpose.
According to the Norwegian Nature Diversity Act, the management goal for species is to ensure viable populations in their natural distribution areas while also attending to their genetic diversity ( §4); to avoid damage to biodiversity ( §6); and apply a precautionary principle when there is little knowledge on the effects of a measure. Our genomic results show that C. bigelowii "subsp. bigelowii" and "C. dacica" are highly separated, supporting a growing body of literature on the recognition of the two taxonomic units belonging to different species. Therefore, seed-sourcing regions must be discussed separately within each taxon. The essential difference between our northern and southern Norwegian clusters, and the geographically similar population clusters reported by Jørgensen et al. (2016), is in the taxonomical hierarchy and subsequent considerations of intraspecific genetic diversity relevant for delimitations of seedsourcing regions. This affects restoration guidance, as seed-sourcing regions will no longer be based on an intraspecific division between two population groups, or between supposedly broadly sympatric subspecies growing in different habitats throughout Norway.
Seed-sourcing regions, seed (transfer) zones, or local provenance areas refer to regions that have similar environments in which natural genetic exchange occurs so that transfer of plant material within the zones should have little or no negative impact (Hufford & Mazer, 2003;Vander Mijnsbrugge et al., 2010). Further, a seed batch is considered appropriate for ecological restoration purposes when its genetic diversity represents its population of origin, and is used on a restoration site of suitable ecological conditions (Erickson & Halford, 2020;Pedrini & Dixon, 2020). Based on the expectation that populations are locally adapted, the idea that the use of so-called local seeds always is the best seed-sourcing strategy is, however, a subject for debate. Other strategies include composite provenancing (regional), admixture provenancing, climate-adjusted, and predictive provenancing to enhance a species' adaptive potential (Breed et al., 2013;Bucharova et al., 2019;Havens et al., 2015;Prober et al., 2015). For a widespread species like Daucus carota (wild carrot), non-local seeds will work just as well as local seeds for restoration (Reiker et al., 2015), while seed-sourcing regions for the endemic Eucalyptus melliodora in Australia have been defined to range 500 km (Supple et al., 2018), the distances separating sites within C. bigelowii "subsp. bigelowii" and "C. dacica." Supple et al. (2018) further suggested that seeds could be sourced more broadly from favorable sites with higher seed quality, increasing genetic diversity without mixing divergent linages. Also, for Eucalyptus cuprea, mixed material was recommended to be used for restoration (Sampson & Byrne, 2016).
Estimates of genetic neighborhood size may, together with our assessments of genetic structure and diversity, inform the choice of donor sites and guide sampling distance for seeds of each of C. bigelowii "subsp. bigelowii" and C. dacica in Norway. The spatial autocorrelation in genetic relatedness varied between the two taxa. Low levels of spatial autocorrelation were found in the northern C. bigelowii "subsp. bigelowii," and a large genetic neighborhood size of up to 300 km was estimated. Apart from interpreted hybrid populations, the sNMF analyses found little genetic structure within C. bigelowii "subsp. bigelowii." Thus, seed sources for C. bigelowii "subsp. bigelowii" in northern Norway located within a 300 km radius can be considered "local" (cf. Norwegian Nature Diversity Act) for ecological restoration purposes. For the southern "C. dacica," a stronger spatial autocorrelation was found, and the genetic neighborhood size was estimated to be 100-150 km. Although a major genetic cluster prevailed, some local genetic distinctiveness not interpreted as hybrids was evident within "C. dacica" (e.g., sites 3 Finse and 7 Memurudalen; Figure 2). As the origin and abundance of these local genetic groups are not known, it would be speculative to interpret them as separate seed-sourcing regions based on our data. However, they do illustrate how using genomic methods to assess overall genetic structure and diversity can identify both larger seed-sourcing regions and specific sites representing distinctive genetic variation.
Our choice of study system was not based on the suitability of C. bigelowii s.lat. for ecological restoration of alpine vegetation per se; although it has been recommended for the purpose, and is a wellknown study species in alpine ecology, it has a partially unresolved taxonomic complexity. We found two distinct evolutionary lineages that can be recognized morphologically, but our results also revealed widespread hybridization, potentially complicating a feasible use of C. bigelowii subsp. bigelowii or C. dacica for ecological restoration.
Within each seed-sourcing region, collecting seeds from fertile hybrids can, as far as possible, be avoided by identifying and recognizing intermediate genetic and/or morphological characteristics.
Matching the habitats of the donor and restoration sites may further contribute to ensure a genetic match, as unknown small-scale genetic differentiation between different habitats is found in many species (Vander Mijnsbrugge et al., 2010).

| Conclusion and future perspectives
Parties to the Convention on Biological Diversity have acknowledged a taxonomic impediment to the sound management of biodiversity, and agreed to identify components of biological diversity important for its conservation and sustainable use (Article 7;CBD, 1992). The use of genomic tools in practical restoration ecology is still in its very beginning (Breed et al., 2019), and here we have shown how population genomics can alleviate the taxonomic impediment by offering a detailed genetic structure of a species complex, and inform practical questions and legal requirements related to improved local seedsourcing. Ideally, genetic and genomic data should be supplemented with common garden or reciprocal transplant experiments to further determine seeds most optimal for each ecological restoration site (De Kort et al., 2014), followed by further genetic assessment and monitoring with measures of genetic resilience of the restored populations (Thomas et al., 2014;Williams et al., 2014). Together with traditional taxonomic and ecological work, genomic information can enable the use of genetically appropriate seeds.  It is a common practice in population genetic analyses to filter out singletons, as they can potentially be contamination, sequencing errors or true rare loci, the latter of which would be considered noninformative in most population genetic analyses. Such loci are filtered out with minor allele frequency (MAF) or by restricting the amount of missing data allowed, although this is debatable since one might also be losing true and valuable information by doing so (Huang & Knowles, 2014;Linck & Battey, 2019). Linck and Battey (2019) found that model-based analyses to infer population structure is strongly affected by the MAF choice, whereas nonmodel-based analyses are less sensitive, and recommend to supplement model-based analyses with nonmodel-based analyses. Potentially wrongly merged loci,

DATA AVA I L A B I L I T Y S TAT E M E N T
where two different loci have been merged to be alleles in one loci, would be expected to occur as heterozygotes throughout the dataset. To exclude such potential loci [--max-obs-het] was set to [0.8].
From the 387 M raw paired reads retrieved in our study, 328 M were retained after demultiplexing and quality controls (see Table  S1 for numbers of raw and retained reads per individual

MAF sensitivity analyses
We ran a sensitivity analysis to illustrate the effect of the minor allele frequency filter (MAF) on the downstream analysis using sNMF (10 independent runs per K for K = 1-10) with MAF values 0.01, 0.05, 0.1, and 0.3. By looking for the "elbow" point on the resulting curves of the cross-entropy criterion, K = 2, K = 3, and K = 4 were chosen to be displayed for the sensitivity analysis.
The first difference was seen in the cross-entropy values ( Figure   S1), but although the values changed, the form of the curve did not change dramatically. Most likely, K (or range of most likely K values) was inferred consistently using different MAF values.
The second difference was seen in the affiliation of individuals to groups in the different runs ( Figures S2-S4 Our choice of MAF value was intended to allow filtering out alleles only occurring in one individual, thus likely being a sequencing error and/or not biologically informative, and to keep rare alleles occurring in more than one individual (thus having an allele count >2, since one individual can be homozygote for one allele).