Introduction

Invertebrates inhabiting the upper intertidal zone of oceanic islands are vulnerable to diverse anthropogenic effects, including overexploitation of species for human consumption. Protective strategies should focus on preserving the viability of threatened conservation units, especially in those exploited species with poorly resolved taxonomy and limited geographic distribution. Conservation initiatives and regulatory frameworks rely on appropriate definitions of distinct conservation units to support implementation and efficient resource allocation (Supple and Shapiro 2018). The taxonomic level of a species does not necessarily coincide with a conservation unit but it is a good starting point to have a robust species delineation, with consensus across various species concepts and data types. When empirical data support them, divergent populations that represent the genetic diversity of a species can be considered evolutionary units for conservation (Coates et al. 2018).

The Macaronesian archipelagos constitute the northeastern Atlantic distribution of the endemic limpet species complex classified as Patella candei d’Orbigny, 1840, which includes four subspecies: P. candei gomesii Drouët (1858) in the Azores; P. candei ordinaria Mabille 1888 in Madeira; P. candei crenata d’Orbigny 1840 in the Canary Islands and P. candei candei d’Orbigny 1840 in the Selvagens and Canary Islands (Christiaens 1973; Hernández-Dorta 1992; Weber and Hawkinks 2002). The majorera limpet, P. candei candei d’Orbigny 1840, is endemic to the islands of Fuerteventura and Selvagens. Currently, these populations represent the global distribution of this limpet, which is highly valued locally for human consumption. The critical status of the majorera limpet population on Fuerteventura in the Canary Islands prompted its inclusion in the Spanish Catalogue of Endangered Species under the category ‘in danger of extinction’ (González-Lorenzo et al. 2015). In contrast, the population on the inhabited archipelago of Selvagens islands is healthy, although the cliff habitat is considered unfavourable for development (Hernández-Dorta 1992; González-Lorenzo et al. 2015).

The Fuerteventuran population has shown a decline in abundance, following a reduction in recolonisation events, loss of connectivity among reproductive populations, low efficacy of conservation strategies and legal protection and uncertainty regarding its taxonomic and historical geographical distribution (González-Lorenzo et al. 2015). The critical conservation status of the P. candei candei population in the Canary Islands requires conclusive data for the design, planning and implementation of a recovery plan, supported by the maximum possible knowledge to guarantee its success. This involves elucidation of the taxonomic position of P. c. candei, estimation of genetic diversity and assessment of intra- and interspecific relationships.

Previous phylogenetic and population genetic studies have not yet clarified the taxonomic position of the populations of P. c. candei and the phylogenetic relationships within this complex. The mitochondrial 12S/16S rRNA molecular phylogeny of the Patella genus confirms close relationships among P. candei, P. lugubris and P. caerulea (Koufopanou et al. 1999). The analysis of independent and combined data sets involving COX1, 12S and 16S rRNA sequences revealed the structure of mitochondrial lineages found among the Macaronesian populations: the paraphyly of P. candei, a clade containing the Azores, Madeira and Desertas samples, and a second clade including individuals from the Selvagens and Canaries and P. lugubris from the Cape Verde Islands (Sá-Pinto et al. 2005). These same lineages were further investigated in detail to elucidate the population structure and gene flow among Macaronesian populations. As a result, mitochondrial data supported two P. candei sub-clades related to P. c. gomesii and P. c. ordinaria subspecies in Azores and Madeira archipelagos, respectively (Sa-Pinto et al. 2008; Carreira et al. 2017). However, the phylogenetic status of P. c. crenata in the Canary Islands and P. c. candei in the Selvagens and Canary Islands remains unclear.

The analysis of nuclear short tandem repeat (STR) allowed to retrieve three clades: (i) Azores, (ii) Madeira and (iii) Gran Canaria in the Canary Islands, in total congruence with expectations from mitochondrial DNA phylogeography (Faria et al. 2017). In addition, shell shape also showed differentiation among populations (Carreira et al. 2017; Faria et al. 2017).

However, to date, no useful information is available to integrate into the evaluation of conservation strategies for P. p. candei since (i) mitochondrial analyses recovered a single clade, without any differentiation between the samples from the Canaries and Selvagens (Sá-Pinto et al. 2005; Sa-Pinto et al. 2008; Carreira et al. 2017) and (ii) STR analysis does not include samples from Selvagens or Fuerteventura, obviating the P. c. candei taxa (Faria et al. 2017).

The current accessibility to genomic information allows to obtain data, useful for conservation, at a resolution, sensitivity and precision not previously achievable (Supple and Shapiro 2018). In particular, the use of genome reduced-representation methodological approaches, e.g. restriction-site associated DNA sequencing (RAD), has enabled researchers to obtain thousands of markers dispersed through the genome in a cost-effective way (Baird et al. 2008; Carleton 2011; Peterson et al. 2012; Kess et al. 2015). This high-resolution data is suitable for reviewing hypotheses previously evaluated with partial mitochondrial data (Sá-Pinto et al. 2005; Sa-Pinto et al. 2008; Carreira et al. 2017). In such cases, the conclusions achieved may be affected by a lack of resolution caused by a recent divergence of species, introgressive hybridization events or other features of the evolutionary dynamics of mitochondrial DNA (Avise et al. 1987; Moritz et al. 1987).

This study focused in (i) the delimitation of the closely related taxa inhabiting the Selvagens and Canary Islands archipelagos (P. c. candei and P. c. crenata), and (ii) the population genetics of a threatened endemic Fuerteventurean limpet (P. c. candei) for assistance in conservation and management. Accordingly, we assessed species delimitation patterns using all available mitochondrial COX1 data and performed a genome scan based on double digest restriction-site associated DNA (ddRAD) to infer an alternative species delimitation and the population structure pattern relevant to the conservation of the majorera limpet.

Materials and methods

Sample collection

The available tissue included frozen (-20ºC) ethanol-preserved samples initially collected in 1995, 2010 and 2014 for mitochondrial DNA analysis of P. candei gomesii Drouet, 1858; P. candei ordinaria Mabille, 1888; P. candei crenata d’Orbigny, 1840; and P. candei candei d’Orbigny, 1840. Additional samples from P. candei crenata and Patella lugubris Gmelin, 1791, were sampled in 2020–2021. The sample subset selected for genome-wide data analysis included P. candei candei samples from Selvagens (N = 30) and P. candei crenata from Fuerteventura (N = 15) and Tenerife (N = 15). Specimens from the P. candei candei threatened population from Fuerteventura (Canaries) were available from (i) a 1995 collection when no restrictions on capture were implemented (N = 5), (ii) the seizure of illegal captures (N = 3) and (iii) non-invasive mucosal swab sampling (N = 2) (Table 1).

Table 1 Sampling locations and data generated for Macaronesian limpet taxa
Full size table

DNA isolation, COX1 amplification and sequencing

Total DNA was isolated with an E.Z.N.A.® Mollusc DNA kit from a muscle tissue sample (10–30 mg) and mucosal swab samples, spectrophotometrically quantified with a Biophotometer Eppendorf D30, electrophoretically characterised and stored at -20ºC.

A fragment (573 bp length) located in the 5’ portion of the COX1 gene was amplified with the PATCOX1F (5’-CCTAGGRATATGRGCAGGTTTAGTAGG-3’) and PATCOX1R (5’-TCAGTTAATARTATAGTAATTGCYCCRGCTA-3’) primers. Reactions were performed in 15 µL volume containing GoTaq Flexi Buffer (Promega, Madison, WI, USA). PCR products were purified and sequenced following previously described protocols (Quinteiro et al. 2020). After chromatogram edition, sequences were aligned using Clustal X (Thompson et al. 1997) implemented in Geneious Prime (http://www.geneious.com).

Elaboration and sequencing of ddRAD libraries

Four reduced representation genomic libraries were elaborated using the ddRAD methodology (Baird et al. 2008; Carleton 2011; Peterson et al. 2012; Kess et al. 2015). In the preliminary methodology set-up, no significant differences among DNA samples replicates (N = 3, 4 samples) were observed (Mastretta-Yanes et al. 2015). A total of 70 DNA samples were processed, to reduce genotyping error, in population specific libraries for P. candei candei, sampled in Selvagens (N = 30, 2014) and Fuerteventura (N = 10, 1995; 2021), and P. candei crenata, collected on Fuerteventura (N = 15, 2020) and Tenerife (N = 15, 2020).

The estimated genome size for Acmaeidae limpet species (Patellogastropoda) (C value = 0.60, 586.8 Mb) (www.genomesize.com) and a draft assembly of transcriptome data (Werner et al. 2013) of Patella vulgata were used to estimate the most appropriate restriction enzyme combination for the selected size range and fragment yield with ddRADseqTools (Mora-Marquez et al. 2017). Total DNA was double digested with High Fidelity (HF®) EcoRI and HindIII restriction enzymes (NEB). The enzyme specific adapters were ligated to the fragments with T4 DNA Ligase (NEB). Adapter-fragments were gel-based size selected (300–500 bp) and purified with Nucleospin (Machery-Nagel). The fragments for each individual were labelled using dual-barcoded index PCR with Phusion™ High-Fidelity DNA Polymerase (Thermo Scientific) and gel purified. The individual libraries were quantified with KAPA Library Quantification Kit (Illumina®) universal qPCR (Roche) to elaborate an equimolar pool. A 12.5 nM library was used for sequencing with MiSeq Reagent kit v2, 2 x 250 bp (Illumina).

Mitochondrial COX1 DNA analysis

A complete dataset was elaborated from 807 available Genbank COX1 sequences from the Patella genus. A phylogenetic neighbor-joining tree was inferred from distances computed using the Tamura-Nei 93 (Tamura and Nei 1993) selected model, with a shape parameter = 0.37 for the gamma distribution of the rate variation among sites, using MEGA7 (Kumar et al. 2016). The p-distances were then estimated within and among those clades/species phylogenetically supported by a bootstrapping value > 70 (2000 replicates). The sequences belonging to the candei/lugubris clades were excluded for the estimation of inter- and intraspecific distance reference values estimated from another Patella spp.

The mitochondrial COX1 phylogenetic inference within the candei/lugubris clade was based on all available GenBank data for the involved taxa and the sequences obtained. Patella caerulea and P. ferruginea were established as outgroups. The data set included 263 taxa with 515 sites. The phylogenetic relationships were analysed by Bayesian analysis with BEAST v2.6 (Drummond et al. 2012) and visualised with DensiTree (Bouckaert and Heled 2014). The previously selected Tamura-Nei 93 model (Tamura and Nei 1993), was implemented in BEAST adding the substitution rate parameter under a gamma distribution and the proportion of invariable sites, with the gamma and invariant sites options, respectively. A Yule tree prior was used, with default values, for tree reconstruction under strict molecular clock. Other options The MCMC chain length was 1 000 000.

Variant calling on ddRAD data

The raw and demultiplexed Illumina data were trimmed for quality and adapter content with Trimmomatic v0.35 (Bolger et al. 2014) and arranged into four population samples: Selvagens and Fuerteventura populations of P. candei candei, and Tenerife and Fuerteventura populations of P. candei crenata. These initial data were submitted to two different and alternative pipelines to minimize the effects of genotyping errors, filtering strategies and algorithms assumptions on the biological conclusions. The Stacks v2.60 pipeline (Catchen et al. 2013) started with an additional trimming with the process radtags v.1.44 program in a subset of 52 specimens with the topmost read number to minimize missing data (< 0.5%). The denovo.pl script was used for sequential running of the ustacks, cstacks, sstacks and populations modules. The ustacks (-m 3, -M 8, N 9) and populations (-p 4, -r 0.8) parameters were selected after evaluation of the number of single nucleotide polymorphisms (SNPs) and genetic differentiation estimates achieved from different values and combinations of m (3), M (2 to 8) and N (= M ± 1), and p (2 to 4), R (0.5–0.7) and r (0.8) (Paris et al. 2017; Diaz-Arce and Rodriguez-Ezpeleta 2019).

The dDocent v2.7.8 pipeline (Puritz et al. 2014) included read merging with the OL paired-end sequencing option in PEAR v0.9.8. CD-HIT (Fu et al. 2012) clustered representative sequences with similarity > 90%, followed by reads mapping to those sequences using the MEM algorithm of BWA (Li and Durbin 2009). Variant calling was performed with Freebayes v1.0.2 (Garrison and Gabor 2012). Finally, the SNP data were filtered using a script (https://github.com/jpuritz/dDocent/blob/master/scripts/dDocent_filters) for sites with missing data (≤ 0.01), quality scores (≥ 30), minor allele count (3), individuals with missing data (> 5%), minimum allele frequency (0.01) and including population specific test for HWE.

Population structure

Using the package SNAPP v2.4.8 (Bryant et al. 2012) within the BEAST2 platform (Bouckaert et al. 2019), a coalescent-based tree inference was conducted in a Bayesian framework directly from biallelic markers without estimating a gene tree for each locus. The Beauti prior settings included α = 2, β = 20 and θ = 0.1. The trees recorder during the MCMC analysis (chain length = 100 000) were visualized using the DensiTree program, reflecting the entire posterior distribution of the species tree. Bitwise distances among individuals were calculated using poppr (1000 bootstrapping replicates) and visualised with a dendrogram constructed with ape R programs (Kamvar et al. 2015; R-Core-Team 2021). Principal component analysis (PCA) of SNP data was performed with SNPRelate v1.16.0 (Zheng et al. 2012).

Genetic structure was also inferred by Bayesian model-based clustering of individuals with Structure v.2.3 (Pritchard et al. 2000) for the assignment of individuals to detected populations. Selected options included the admixture model for the ancestry estimates among individuals, the sampling locations as prior information (LOCPRIOR option) and the correlated allele frequencies model. The length of burn-in period was 200 000 and the number of MCMC was 2 000 000. The parameter K = 1 to K = 5 was sequentially evaluated with Structure Harvester v0.6.94 using 5 replicates (Earl and vonHoldt 2011) for the complete and taxon-specific datasets being duplicated for the alternative stacks and dDocent SNP datasets. Then, the Evanno method (Evanno et al. 2005) was used for detecting the number of K groups that best fit these datasets.

Mitochondrial sequences obtained were deposited in GenBank/EMBL/DDBJ and are available with accession numbers MZ605952 - MZ606134. VCF files containing filtered SNP data and MEGA and BEATS input alignment COX1 files were deposited in Dryad (https://doi.org/10.5061/dryad.b5mkkwhfn).

Results

Analysis of genetic distances based on COX 1 mitochondrial sequences

Analysis of the available mitochondrial COX1 sequences belonging to the Patella genus, from this work and mined from GenBank (accessed 03/11/2019), enabled a comparison of the pairwise distance values at different taxonomical levels and among consistent mitochondrial clades within the P. candei complex (Sá-Pinto et al. 2005; Sa-Pinto et al. 2008). The mean net distance between pairs of Patella species, excluding those of the P. candei complex and P. lugubris, was 0.112 (SD = 0.015). A lower estimate, 0.041 (SD = 0.024), was obtained evaluating the inter-clade distances within the P. candei complex, including P. lugubris. The distance between clade A (containing P. c. gomesii and P. c. ordinaria) and clade B (including P. c. candei, P. c. crenata and P. lugubris) was 0.05. The mean estimated intraspecific (species or clade) value was similar for all groups (0.007; SD = 0.005). A lower distance estimation (0.0125; SD = 0.004) was obtained from a comparison between clades A.1 (Azores, P. c. gomesii) and A.2 (Madeira, P. c. ordinaria), in the range of intraspecific range of genetic distances as estimated for the Patella genus (Fig. 1; Table 2).

Fig. 1
figure 1

Estimation of genetic distances (p-distances) based on mitochondrial COX1 gene sequences at different taxonomic levels and detected clustering patterns within the Patella genus and the P. candei complex, including P. lugubris

Full size image
Table 2 Estimation of distances (average p-distance) among clades (below diagonal, blue), within clades (diagonal, green), and SD values (above diagonal, orange)
Full size table

Phylogenetic relationships based on COX 1 mitochondrial sequences

The complete set of available COX1 sequences for the P. candei/P. lugubris-related taxa were consistently grouped into two major clades (bootstrapping value = 99), labelled A and B. Clade A was subdivided into two closer subclades, A.1 and A.2. Clade A.1 included all the sequences from individuals sampled at the Azores, grouped together with data mined from GenBank and related to individuals of the same geographic origin. The sister clade included individuals from the Madeira archipelago: Madeira, Porto Santo and Desertas (Fig. 2). Clade B.1 included specimens of P. candei candei sampled at the Selvagens and Canary Islands. No subclade organization or subdivision related to taxa or sampling site was supported within this broad clade. The close sister clade B.2 included all P. lugubris samples analysed from the distant Cabo Verde archipelago.

Fig. 2
figure 2

Density tree based on all available mitochondrial COX 1 sequences, from this study and mined from GenBank, reflecting the relationships among taxa included in the Patella candei complex, including P. c. gomesii, P. c. ordinaria, P. c. candei and P. c. crenata from Macaronesia, and P. lugubris from Cabo Verde. The specimens sequenced in this study are traced to the sampling site (not drawn to scale) by grey lines. Grey vertical lines represent 0.005 p-distance intervals. The P. caurelea sequences located in the upper position (no label) were used as an outgroup. An embedded map displays the four sampling locations in the northeastern Atlantic

Full size image

ddRAD variants

A total of Illumina 17 312 424 paired-end reads was generated from 70 individuals. The use of both recent and carefully preserved samples, with adequate DNA integrity and concentration (mean > 800 ng), in combination with long-time archived and swab samples, resulted in different sequencing output patterns. Thus, both initial mean reads number and depth coverage per specimen were lower for P. c. candei from Fuerteventura (Pcan FV) (162369.6 and 9.48x mean values) and P. c. candei from Selvagens (Pcan SV) (177892.1 and 11.1x) than those achieved for P. c. crenata (Pcre FV) (346026.6 and 15.2x) and P. c. crenata from Tenerife (Pcre TF) (226757.4 and 13.5x).

dDocent assembled 19 366 sequences into 8494 contigs, a number close to the theoretically expected ddRAD fragments. After initial SNP calling and filtering, 3092 SNPs from 54 individuals were retained. To take into consideration the effect of alternative data on achieved inferences, a highly stringent, with minimum missing data (< 5%), Stacks data set included 46 specimens and 2136 SNPs.

Genetic diversity and differentiation

The estimated nucleotide diversity values for candei were higher than for the crenata taxon, with larger values estimated for the small and threatened Fuerteventura population (Table 3).

Table 3 Number of alleles, loci and nucleotide diversity (nuc. div.) estimates for the two SNP data sets obtained using Stacks and dDocent pipelines in P. candei candei (Pcan) and P. candei crenata (Pcre) sampled on Fuerteventura (FV), Tenerife (TF) and the Selvagens Islands (SV)
Full size table

The results of the exact G test in both datasets, for genotype differentiation among populations, suggested a differentiation for all pairwise comparisons (P < 0.0001), except in the case of crenata populations.

The population pairwise F ST from both datasets suggested a strong differentiation between the samples classified as P. candei candei (Pcan) and P. candei crenata (Pcre) taxa (F ST = 0.5, 0.7; P = 0). Within P. candei candei the populations from the islands of Fuerteventura and Selvagens were significantly differentiated (F ST = 0.05, 0.10; P = 0). Also, Tenerife and Fuerteventura populations of P. candei crenata showed a low but significant level of differentiation. The values estimated from both datasets were correlated, but those estimated based on Stacks pipeline SNPs data were generally higher (Table 4).

Table 4 F ST pairwise values among populations of P. candei candei (Pcan) and P. candei crenata (Pcre) sampled on Fuerteventura (FV), Tenerife (TF) and the Selvagens Islands (SV). Estimates based on the SNP dataset derived from the Stacks pipeline are below the diagonal, whereas those from the dDocent pipeline are above the diagonal
Full size table

The phylogenetic relationships among individual genotypes suggested the monophyletic and divergent clustering into two main taxa-specific clades for P. c. candei and P. c. crenata. These clades were well supported and recovered under the Bayesian criteria (Fig. 3. A, B) and distance clustering (Fig. 3. C, D), as well as when alternative Stacks (Fig. 3. A, C) and dDocent (Fig. 3. B, D) data sets were considered. A high level of intraspecific cluster differentiation was observed in P. c. candei respect to the P. c. crenata (Fig. 3. A, B) density trees. Population-specific clades were only well supported (> 93%) for the Fuerteventura and Selvagens populations of P. c. candei in distance trees based in the dDocent data set (Fig. 3. D) whereas that a low support (< 50%) was estimated in Stacks data set based trees. A similar pattern is observed for population-specific clades in P. c. crenata. The population specific clustering observed in trees derived from dDocent data set, however, showed a low support (Fig. 3. D), whereas no clustering was observed from the Stacks data set for P. c. crenata. (Fig. 3. C).

Fig. 3
figure 3

Trees estimated for limpets belonging to the Fuerteventura and Selvagens populations of Patella candei candei (candei) and to the Fuerteventura and Tenerife populations of Patella candei crenata (crenata). Both density trees based on Bayesian criteria (A and B) as dendrograms based in bitwise distance (C and D) were estimated from Stacks (A and C) and dDocent SNPs (B and D) data sets. Vertical lines in density trees represent a 0.01 distance value

Full size image

The PCA revealed a clear divergence between the candei and the crenata populations, along the first axis and explained 26.4–27.6% of the total variance. Furthermore, two different patterns emerged when variation within these taxa was considered. The crenata populations (Fuerteventura and Tenerife) showed a homogeneous and near compact cluster. Conversely, the candei populations, from Fuerteventura and Selvagens, displayed an insular-specific divergent clustering. The second axis principally separated these candei samples with 3.0–3.4% of the total variance explained. Similar patterns were obtained from the alternative Stacks and dDocent LD filtered data sets (Fig. 4).

Fig. 4
figure 4

PCA plot of the first two component axes calculated from a Stacks pipeline dataset (green colour; 1384 LD filtered SNPs) and dDocent data set (blue colour; 2756 LD filtered SNPs) of SNPs from insular populations of P. candei candei (Fuerteventura and Selvagens) and P. candei crenata (Fuerteventura and Tenerife)

Full size image

The Structure model-based clustering method for inferring population structure suggested the presence of three differentiated populations based on diverse analysis of both the Stacks and dDocent data sets.

For the complete data set, involving all populations, a K = 2 was selected, where individuals from P. c. candei and P. c. crenata taxa were separately clustered. However, the test of a K = 3, with a slightly lower likelihood value in Structure Harvest, suggested a population differentiation within the main P. c. candei cluster between Selvagens and Fuerteventura samples. The individual membership coefficient for Fuerteventura individuals shows a different pattern respect to the Selvagens individuals including a significant memberships coefficient for an alternative cluster. In contrast, individuals from Tenerife and Fuerteventura populations of P. c. crenata showed identical and homogeneous ancestry estimates (Fig. 5. A, B).

Fig. 5
figure 5

Summary plot of estimates of ancestry (Q). Plots were divided into population specific segments (Fuerteventura, Selvagens and Tenerife). Each individual is represented by a single vertical line broken into K coloured segments, with lengths proportional to each of the K inferred clusters. The plots for the complete combined data set (A and B) (P. candei candei and P. candei crenata) assumed K = 3 in A (Stacks data set) and B (dDocent data set). In addition, plots were obtained separately for each taxa assuming K = 2. The P. c. candei results were separately plotted (left) from Stacks (C) and dDocent (D) data sets. Similarly, the specific P. c. crenata estimates were plotted (right) from Stacks (C) and dDocent (D) data sets

Full size image

In a subsequent analysis, the P. c. candei and P. c. crenata taxa were discretely analysed. The Structure Harvest results suggested a K value of 2 and the analysis revealed a partition of P. c. candei into two clusters, for both Stacks and dDoc SNPs data sets, in agreement with the Selvagens and Fuerteventura geographic origin of individuals. The analysis (K = 2) of the P. c. crenata based on the Stacks SNPs data set reflected a homogeneous membership coefficient for the sampled individuals. In contrast, when ancestry was evaluated with the dDoc data set, a different proportion of membership of each P. c. crenata samples from Tenerife and Fuerteventura was obtained (Fig. 5. C, D).

Discussion

Analysis of the ddRAD-based SNPs data provided the resolution required to infer taxonomic and population relationships at the level where mitochondrial data showed critical limitations. Moreover, the data achieved from archived samples and non-invasive mucosal swab sampling suggest the efficacy of this protocol under non-ideal conditions. The subspecies P. c. candei and P. c. crenata were genetically differentiated, for the first time, based on the diverse SNPs datasets and alternative analysis. At the intraspecific level, differentiation has also been observed between the Selvagens and Fuerteventura populations of P. c. candei, which represents a significant contribution to the planning of conservation strategies for the endangered population.

Disjunct ranges for inter- and intraspecific genetic distances are essential for drawing species boundaries under the COX1 DNA barcoding concept, allowing for the molecular identification of known species and the discovery of previously disregarded taxa (Hebert et al. 2003). An evaluation of estimated genetic distances is particularly pertinent in the absence of a taxonomic consensus involving threatened populations. The mean interspecific COX1-based distance among Patella species, excluding candei and lugubris clades, was 11.2%. The mean intraspecific or intra-clade distance among Patella species (0.7%) agreed with expected values of > 1%, as reported for Canadian marine molluscs (0.49%) (Zou et al. 2011; Layton et al. 2014). The distance among the taxa of the candei complex and P. lugubris is at an intermediate level, with a mean value of 4.1%. However, there was no divergence between the putative taxa P. candei crenata and P. candei candei.

Mitochondrial-based phylogenetic studies of Macaronesian species of the genus Patella revealed highly consistent patterns (Koufopanou et al. 1999; Sá-Pinto et al. 2005; Sa-Pinto et al. 2008; Carreira et al. 2017). In congruence, our analysis grouped P. candei and P. lugubris as the closest taxa within the clade I (Sá-Pinto et al. 2005), denominated the P. candei–P. lugubris complex (Sa-Pinto et al. 2008). Within clade I, clade MAZ, following Sá-Pinto et al. (2008) nomenclature, was defined here as clade A.

The division of clade A into the subclades A.1 and A.2, with individuals from the Azores and Madeira archipelagos, respectively, reflects the taxonomic delineation of P. c. gomesii (Azores) and P. c. ordinaria (Madeira), elevated to the species range (Faria et al. 2017). Similarly, clade B (clade CANCBV following Sá-Pinto et al. (2008) nomenclature is subdivided into clade B.1, containing P. candei candei individuals from the Selvagens and Canary archipelagos, and clade B.2, including P. lugubris.

Clade B1 includes two putative and partially sympatric subspecies P. c. candei and P. c. crenata (Christiaens 1973). No genetic discontinuity and monophyletic clustering were observed within this clade, here and in other studies (Carreira et al. 2017) from mitochondrial data. In contrast, their species status has been assumed in a STR-based study although, unfortunately, no STR data have been obtained from P. candei candei taxa (Faria et al. 2017). However, morphological differentiation has been described between candei and crenata shells based on the position and shape of the apex and radiation ridges and, notably, a smooth edge without ribs in candei (Christiaens 1973; Hernández-Dorta 1992; Titselaar 1998; González-Lorenzo et al. 2015). This description resembled the eroded phenotype described in shells of P. c. candei from Selvagens (Carreira et al. 2017). Despite the non-cryptic appearance, the absence of previous mitochondrial or nuclear data supporting P. c. candei and P. c. crenata clades resulted in unresolved phylogenetic relationships and in the absence of a robust taxonomic frame required for the conservation and management of the putative candei and crenata taxa.

The subspecies Patella c. candei and P. c. crenata individuals were divergent and monophyletically clustered in well-supported clades (100%). The F ST pairwise values among populations from different taxa (F ST > 0.5) suggested an absence of gene flow, with high values when comparing those populations sympatrically occurring in the Fuerteventura rockeries. Finally, most of the total variance estimated from the SNP data was related to differences in allelic frequencies between P. c. candei and P. c. crenata.

Diverse events have been suggested as a cause of incongruence between nuclear and mitochondrial data. Thus, introgressive hybridization events introduce female mtDNA lineages of a species in the genomic composition of a second species. Such a case of putative hybridization between P. depressa and P. vulgata has been reported (Koufopanou et al. 1999). However, in the P. candei complex the two main mitochondrial lineages, clades A and B, diverged at intermediate values between intra- and interspecific rates, suggesting a recent speciation process within the complex and an incomplete mitochondrial lineage sorting affecting the closely related candei and crenata taxa.

The SNP data suggested the delimitation of candei and crenata sister taxa as divergent evolutionary units. A definitive conclusion on their taxonomic status, however, requires a phylogenetic analysis of homologous data involving all taxa and populations within the candei complex and related species. In any taxonomic context, the observed delimitation and high divergence between these morphologically similar taxa must be considered for the adoption of problematic specific management programmes for a resource inhabiting the same rockeries on the Fuerteventura coast (Titselaar 1998; González-Lorenzo et al. 2015).

The resolution provided by the SNP data enabled the observation of a different population structure pattern within the candei and crenata taxa. The populations of crenata sampled from Tenerife and Fuerteventura belong to a single gene pool. Only when using the wider SNPs dataset was a slight population differentiation observed.

In the case of candei, the isolation between the only two worldwide populations of P. c. candei enables us to delineate intraspecific units for conservation. The significant genetic differentiation between the Selvagens and Fuerteventura samples reflects an absence of gene flow congruent with the absence of natural recruitment in Fuerteventura from a putative source population is Selvagens (González-Lorenzo et al. 2015). The pattern observed for P. c. candei resembles that detected for the endangered limpet P. ferruginea. Throughout the Mediterranean, non-differentiated mtDNA lineages were detected. In contrast, northwestern local populations were differentiated by ISSR nuclear markers analysis, with implications for the planning of conservation initiatives (Casu et al. 2006, 2011).

As stated by the Convention of Biological Diversity, intraspecific diversity is a fundamental part of overall biological diversity (CBD 1992) since the conservation of distinct local populations is vital for maximizing evolutionary potential and minimizing the probability of extinction (Allendorf et al. 2013). The restoration plan for the “majorera limpet” suggests the possibility of restoring the threatened population of Fuerteventura with the Selvagens population as a source. However, this possibility would only be considered after the analysis and evaluation of resolutive genetic data. In consequence, this study responded to the Plan proposal for an action involving the genetic characterization of the “majorera limpet” and the analysis of inter-population gene flow (Gobierno de Canarias 2015).

Source populations for translocations should have genetic similarity to the recipient population and high genetic diversity (Allendorf et al. 2013). The SNP data acquired from a reduced representation of the genome of P. c. candei suggested a genetic differentiation between samples from the Selvagens and Fuerteventura co-specific populations. The introduction of individuals from Selvagens carrying differentiated genotypes into the threatened Fuerteventura populations may have hardly predictable consequences.

The genetic diversity estimated (N alleles = 18) for the near-to-collapse population in Fuerteventura population is unexpectedly slightly higher or similar to that estimated (N alleles = 24/38) for the healthy unexploited Selvagens population. Nevertheless, erosion of the current genetic diversity in the Fuerteventura population would be expected after supplemental translocations under conditions of climate change and habitat degradation.

The genetic diversity values do not support the role suggested for the Selvagens population of P. c. candei as the ancestral population of P. c. candei, or the ancestral centre of species radiation (Weber and Hawkinks 2002; Sa-Pinto et al. 2008) if we consider its relationship with Patella lugubris based on mitochondrial phylogenies (Sá-Pinto et al. 2005; Sa-Pinto et al. 2008). In contrast, both Selvagens and Fuerteventura likely constitute relic populations, healthy and threatened, respectively. Selvagens was likely peripherally differentiated, and the Fuerteventura population partially retains the genetic diversity of a pre-human ancestral distribution that existed before excessive exploitation (Weber and Hawkinks 2002). However, González-Lorenzo (2015) rejected the relictual hypothesis based on the archaemalacological register derived from deposits of aboriginal origin where P. candei was abundant only in Fuerteventura and Lanzarote. The relictual hypothesis will be congruent with the archaeological register only if natural factors are responsible for the current distribution. Precise molecular dating, which is currently unavailable, will aid in clarifying this theory.

Without any inference about the evolutionary forces shaping the detected genetic variability in P. c. candei, the observed patterns are valuable and unique evidence to support urgent decisions on the conservation management and strategy for the threatened Fuerteventura population. We conclude that the “majorera limpet” is a significant evolutionary unit requiring closed and population-specific management focused on the preservation of exceptional genetic diversity with which to face future environmental challenges.