For this study, 28 Secale cereale L. subps. cereale accessions and its crop wild relative (S. strictum subps. strictum, referred as S. strictum hereforth) were selected from different geographic regions. The panel consist of eight ex situ accessions from gene banks, nine cultivars (“Imperial,” “Kungs II,” “Petkus,” “Dankowskie Zlote,” “Ailé,” “Voima,” “Alvão,” “Pulawskie,” and “Antoninskie”), along with 11 in situ accessions collected “on farm” in 2014 from Northeast Portugal (Table 1). Ex situ accessions were provided by the following gene banks with acronym, accession prefix and country: Leibniz-Institut für Pflanzengenetik und Kulturpflanzenforschung (IPK, R, Germany) and Nordic Genetic Resource Center (NordGen, NGB, Sweden). Rye accessions were grouped and mentioned hereforth as cultivars, ex situ, and in situ/on farm accessions. Cultivars are those resulting from modern rye breeding, in situ are collections held in farmers' fields and ex situ are those collections detained in gene banks or botanical gardens. Both ex situ and in situ accessions collectively will be referred as landraces. The 11 in situ Portuguese accessions were collected from local farms on Northern region (Figure 1A) where a great importance of rye culture still exists at regional level. The region sampled covers part of the Serra da Estrela Natural Park, the largest natural conservation area, and biggest mountain range in Portugal, with several valleys along the mountain assortment. Such region is characterized by harsh winters and samples selected are within 400–1100 m elevation (Figure 1B). Each accession represents a mixed sampling from small plots (maximum 0.5 ha) maintained by local farmers under a subsistence regime. The samples collected were sowed in September/October 2013 and harvested in mid-July 2014, as it is commonly performed. About 200 kg from each in situ population were obtained directly from each farmer and further used on genetic diversity studies and stored at the Instituto Superior de Agronomia, University of Lisbon for future characterizations of agronomical and morphological traits. All in situ accessions included in this study are available upon author request.

DNA from 5 to 10 individuals of each rye germplasm accessions (cultivars and ex situ), while for Portuguese in situ collections 16 individuals were used, for accounting within-population diversity. Genomic DNA (gDNA) was extracted from young leaves using the cetyltrimethylammonium bromide (CTAB) procedure adapted from Thomas et al. (1993). Briefly, young leaves (100 mg) from individual plants grown from seeds were directly ground by an Eppendorf-pestle in a 1.5 mL tube, thawed and resuspended in 300 μl of extraction buffer [0.35 M Sorbitol, 0.1 M Tris pH 8.0 NaCl, 5 mM EDTA, 1 % (w/v) PVP-40] by adding 3.8 g/L of sodium bisulphite and 1 % (v/v) 2-mercaptoethanol (Sigma Aldrich) upon use. After a 5-min incubation on ice, 300 μl of lysis buffer [0.2 M Tris pH 8.0, 2 M NaCl, 50 mM EDTA, 2% CTAB] was added along with 120 μl of 5% sarkozyl (w/v) and RNAse A (10 mg/mL). After incubation at 65°C for 15 min, an equal volume of chloroform: isoamyl alcohol (24:1) was then mixed by a brief vortex, and the aqueous phase was recovered by centrifugation at 16,000 g for 15 min. gDNA was precipitated with 0.6 volume of isopropanol and recovered by a 15-min centrifugation at 16,000 g, followed by a washing step with 70% ethanol. Following a centrifugation at 16,000 g for 10 min, the pellet was dried and resuspended in 40 μl deionized water. DNA purity and concentration were measured at 260/280 nm using a spectrophotometer (NanoDrop-1000, Thermo Scientific) while DNA integrity was verified by agarose gel electrophoresis.

A set of 16 microsatellite markers (Supplementary Table S1) was used for screening rye accessions, consisting of nine genomic SSRs (gSSRs, Saal and Wricke, 1999) and seven Expressed Sequence Tags-SSRs (ESTs-SSRs, Hackauf and Wehling, 2002). Before multiplexing, each SSR marker was validated in singleplex polymerase chain reactions (PCR) using a three-primer PCR approach (sensu Schuelke, 2000) for reaction reproducibility and presence of PCR artifacts. Each SSR was PCR amplified in a 25 μl volume reaction following cycling conditions previously described (Saal and Wricke, 1999; Hackauf and Wehling, 2002), using HotStarTaq DNA Polymerase kit (Qiagen), as per manufacturer's instructions. After, SSRs amplified fragments were run in an ABI 3130XL sequencer (Applied Biosystems) with the internal size standard GS500 LIZ (Applied Biosystems), while allele calling was performed in GeneMapper v 3.7 (Applied Biosystems). Stringent selection of markers to ensure the success of co-amplification loci using Multiplex Manager software (Holleley and Geerts, 2009), allowed building four SSRs panels assembled in 4-plex PCR reactions (Multiplex A, B, C, and D; Table 2), using four universal forward fluorescently labeled primers following Culley et al. (2013). To increase genotyping accuracy, a “Pig-tail” sequence was added at the 5′ end of each of the reverse primer (Brownstein et al., 1996). PCR multiplex amplifications were carried out using the QIAGEN Multiplex PCR kit (Qiagen), following the manufacturer' s protocol, in a total volume of 25 μL with 1 μL of gDNA (50–100 ηg) and 2.5 ρmol of each primer Forward and Reverse and 0.15 ρmol of each of the tailed fluorescently labeled primers (D1–D4). Reactions were done in 96 well-plates and on each plate one sample was repeated per run thus working as positive control for scoring. Negative PCR controls were included. Initially, a hot-start step at 95°C for 15 min was performed, followed by a touchdown cycling protocol adapted from Hackauf and Wehling (2002): 5 cycles of denaturation at 95°C for 45 s, primer annealing at 68°C for 5 min with −2°C/cycle; a sequence extension at 72°C for 1 min; 5 cycles of denaturation at 95°C for 45 s, primer annealing at 58°C for 2 min with −2°C/cycle and an extension step for 1 min at 72°C; 27 cycles at 95°C for 45 s, 47°C for 75 s, and 72°C for 1 min; followed by a final extension step at 72°C for 10 min. Multiplex PCR products were run as described earlier and SSR allele sizes were aligned with the internal size standard and scored using the binning function in GeneMapper v3.7. (Applied Biosystems). To improve the SSR marker data quality, allele assignments were checked manually, and ambiguous results were set as “missing data.”

Genotyping errors were assessed using MICRO-CHECKER v2.2.3. (Van Oosterhout et al., 2004), and estimation of null alleles frequency was done with the EM algorithm of Dempster et al. (1977) as implemented in FreeNA (http://www.montpellier.inra.fr/URLB/). These values were computed, as described in Chapuis and Estoup (2007), with 10,000 bootstrap iterations, alternatively using and not using the Excluding Null Alleles (ENA) method, after assessment of null allele frequencies. Polymorphism information content (PIC) and genetic diversity indices were calculated with Microsatellite Toolkit v.3.1.1 (Park, 2001) and GenALEx 6.5 (Peakall and Smouse, 2006, 2012). These included the total allele number and mean alleles per locus (N_a), private alleles, inbreeding coefficient (fixation index, F), observed (H_O), and expected (H_E) heterozygosity. Deviations from Hardy–Weinberg equilibrium (HWE) were assessed for each locus-population combination and linkage disequilibrium (LD) to determine the extent of distortion from independent segregation of loci using GenePop v4.5 (Rousset, 2008). Statistical significance for both HWE and LD was tested by running a Monte Carlo Markov Chain (MCMC) consisting of 10,000 iterations each, and p-values were corrected for multiple comparisons [p < 0.00012, (0.05/406)] by applying a sequential Bonferroni correction (Rice, 1989).

To detect isolation-by-distance (IBD) effects, F_ST/(1 – F_ST) and FSTENA/(1 – FSTENA) matrixes were done, with a geographic distance matrix defined as pairwise distances generated from geographical coordinates expressed in Km. Pairwise unbiased F_ST-values using the ENA method (FSTENA) for each population comparison were calculated with FreeNA software, while F_ST- values were generated in GenALEx 6.5. Both Slatkin's linearized F_ST (F_ST/(1 − F_ST)) matrixes were obtained in GenoDive 2.0b27 (Meirmans and Van Tienderen, 2004). The correlation between the two data matrices was assessed using a Mantel test and its significance estimated by p-values, the regression coefficient (R²), and the mean correlation coefficient (R_XY) over 999 random permutations as implemented in GenALEx 6.5.

Population structure was addressed using three approaches: (i) estimating relations among populations using genetic distances; (ii) hierarchical genetic analysis by AMOVA; and (iii) individual-based clustering.

All 16 SSRs were tested in singleplex reactions at the estimated optimal annealing temperature, and only after this initial quality assessment, SSRs markers were grouped into 4-plex reactions (Supplementary Table S1). Upon multiplex reactions, SCM180 (Multiplex C) displayed a difficult allele scoring performance, not depicted in singleplex reactions. This fact may be ascribed to PCR dynamics under a multiplex reaction, where concentrations of different primers are equimolar, requiring in some cases, relative balanced primers concentrations (Sint et al., 2012). For the remaining 15 SSRs loci, allele profiles were clear and easy to score. No errors in the genotypic data matrix were detected, indicating the absence of potential errors associated with stuttering bands or large allele dropout in SSRs screened. In only 10 of the 406 locus-populations comparisons, the frequencies of null alleles were higher than 0.20 (Table 2). Deviations from Hardy–Weinberg Equilibrium (HWE) were observed in most loci except for SCM166, with 70 locus-population combinations statistically significance (p < 0.05); while after sequential Bonferroni correction only four loci (SCM64, SCM39, SCM63, SCM75) displayed significant deviations, matching 11 of the 406 locus-population combinations (Supplementary Table S2). All 15 loci were in linkage equilibrium after Bonferroni correction, thus being non-correlated, and alleles independently segregated and inherited (data not shown). Negative fixation index (F) estimates were observed across several loci, exceptions for the EST-SSRs SCM152 (0.060), and SCM63 (0.017) and in the gSSR SCM39 (0.057, Table 2), which can reflect more heterozygotes than expected or other population structure complexities.

Overall, a total of 122 alleles were detected in the 285 individuals analyzed (Table 2). All loci screened were polymorphic except SCM113, which revealed to be monomorphic (194 bp, Supplementary Table S1) being not used for further analysis. The total number alleles per locus ranged from 3 (SCM66 and SCM98) to 13 (SCM28) with an average of 8 alleles per locus (Table 2). When comparing gSSRs with EST-SSRs, the first revealed a higher number (5–13) of alleles per locus than EST-SSRs (3–12), with an average of 9.6 and 7.3 alleles, respectively (Table 2). Once the dataset was separated in cultivars and landraces (including in situ and ex situ accessions) accessions, results showed that in landraces the mean allele number is higher with 4.4 in all SSRs, 4.8 with gSSRs, and 4.4 in EST-SSRs, against the values in cultivars (3.3 alleles per locus in all SSRs, 3.7 in gSSRs, and 3.3 for EST-SSRs, Table 3). When analyzing landrace dataset, ex situ collections (2.97 all SSRs, 3.3 for gSSRs, 2.97 for EST-SSRs, Table 3) showed a substantially lower mean of alleles per locus when compared to in situ accessions (5.5 all SSRs, 6 for gSSRs, 5.5 for EST-SSRs), but similar values to those obtained for cultivars.

Overall, Polymorphic Information Content (PIC) values ranged from 0.39 (SCM66) to 0.85 (SCM138) with a mean value of 0.67 (Table 2). Average PIC-values were higher in gSSRs (PIC_gSSRs = 0.73) than in EST-SSRs (PIC_EST-SSRs = 0.59). Similar PIC-values were observed in cultivars, ranging from 0.366 (SCM63) to 0.695 (SCM138), and in landrace accessions (SCM166: PIC = 0.346; SCM138: PIC = 0.677). In both cultivars and landraces, gSSRs revealed to be more polymorphic than EST-SSRs. In our 14-loci dataset, observed heterozygosity (H_O) varied from 0.51 (SCM166) to 0.85 (SCM86) with a mean of 0.68 (Table 2); and expected heterozygosity (He) varied between 0.480 (SCM166) and 0.86 (SCM138). When narrowing analysis to gSSRs and EST-SSRs, both mean H_O and He were higher in genomic (H_O = 0.73; He = 0.76) compared to the expressed loci (H_O = 0.62; He = 0.64).

The Fixation Index F (also called the Inbreeding Coefficient) exhibits values from −1 to +1. Values close to zero are expected under random mating, while substantial positive values indicate inbreeding or undetected null alleles. Negative values denote excess of heterozygosity, due to negative assortative mating, or selection for heterozygotes. Overall, negative F-values were observed across most accessions (Table 3), thus revealing an outbreeding scenario with increased number of heterozygotes. Positive F was encountered in six Portuguese in situ accessions (SECCE1–SECCE6), ranging from 0.002 (SECCE5) to 0.049 (SECCE1; Supplementary Table S3). Despite positive, their relatively small F-values indicate that these populations are at or near Hardy–Weinberg equilibrium, further supported by the lower observed heterozygosity values against the expected under HWE (Supplementary Table S3).

Pairwise Wright's F-statistics (F_ST) was used as a measure of the extent of genetic differentiation among subpopulations, with values ranging from 0 (no differentiation) to 1 (high differentiation). Mean pairwise F_ST was 0.136 (min = 0.017, “Pulawskie” and “Dankowskie”; max = 0.418, “Riodeva” and S. strictum) indicating an overall low level of population's differentiation (Supplementary Table S4). Most pairwise populations (749 out of 784) showed low to moderate genetic differentiation (pairwise F_ST < 0.250), while the remaining 35 pairwise populations displayed high genetic differentiation (pairwise F_ST ranged from 0.251 to 0.418). The presence of null alleles has not caused a significant overestimation of the level of population differentiation, as low population differentiation was also depicted with FSTENA (mean = 0.138; min = 0.015, “Pulawskie” and “Dankowskie”; max = 0.417, “Riodeva” and S. strictum), with 747 pairwise populations showing low to moderate genetic differentiation (Supplementary Table S4). The remaining 37 pairwise comparisons exhibited high genetic differentiation (pairwise FSTENA > 0.250). Overall, cultivars (mean F_ST = 0.136; FSTENA=0.138) and ex situ collections (mean F_ST = 0.163; FSTENA=0.165) presented a moderate population differentiation, whereas in situ accessions almost no genetic differentiation were depicted (mean F_ST = 0.057; FSTENA=0.059). An overestimation of F_ST-values due to null alleles is widely known especially in cases of significant population differentiation, which is not the case in our study where weak population differentiation was detected.

In order to understand whether genetic variation is correlated with geographical gradients, Isolation-by-distance (IBD) effects were addressed. F_ST-values (Figure 2A either excluding, FSTENA, or not null alleles F_ST Figure 2B) confirmed a small, yet significant, explanation of genetic diversity variation across a geographic range.

Twenty-nine rye accessions were first genotyped with 16 SSRs, and after SSRs quality assessment 14 markers were further used for subsequent genetic diversity analysis (Supplementary Table S6). One SSR was discarded due to low reproducibility in a multiplex PCR amplification (SCM180) and other revealed to be monomorphic (SCM113) across the germplasm analyzed and thus not being an informative marker for posterior diversity analysis. Unexpectedly, no private alleles were detected, contrasting with other SSRs studies in rye populations (Parat et al., 2016; Targońska et al., 2016). In this context, by not detecting private alleles, one can speculate that alleles are present in one or more rye populations regardless of being cultivars or landraces thus representing significant shared allelic diversity. Observed heterozygosity (H_O) obtained in our study (0.51, SCM166–0.85, SCM86) with a mean of 0.68, is higher than those reported for 9 Latvian rye accessions using 9 genomic loci (Gailīte et al., 2013) with a mean H_O of 0.58, ranging from 0.21 (SCM2) to 0.84 (SCM9). Furthermore, a recent study using 32 SSRs in 14 rye accessions (Parat et al., 2016), reported values of H_O per SSR of 0.44 ± 0.17 and 0.67 ± 0.14 for He; while in our study, despite using only 14 SSRs both H_O and He were similar (0.64 ± 0.21 and 0.66 ± 0.22, respectively) likely due to a higher number of rye populations screened (n = 29).

PIC-values obtained for SSRs used were high (average PIC = 0.67) which indicates their high informativeness, which was predictable since the SSRs markers were not selected randomly, but based on the previous performance analyses (Shang et al., 2006). Our PIC-values were higher when compared with a recent study using 22 SSRs (including EST-SSRs and gSSRs) to screen 367 Polish rye accessions, which displayed a 0.57 average PIC-value for all loci used (Targońska et al., 2016); but similar to those obtained in 14 rye accessions using 32 SSRs (PIC-values ranged from 0 to 0.92) with an average of 0.62 (Parat et al., 2016). We detected higher PIC-values in gSSRs (0.73) than in EST-SSRs (0.59), which is in agreement with a previous study using 8 gSSRs (PIC = 0.63) and 14 EST-SSRs (PIC = 0.54; Targońska et al., 2016). Inversely, another analysis of rye cultivars using both gSSRs and EST-SSRs, showed a lower polymorphism content in 13 genomic SSRs (0.38) than in 11 EST-SSRs (0.58; Shang et al., 2006). Deviations on PIC-values depicted in our study might be due to differences in plant material sources compared to former reports, which may influence the number of alleles detected at each SSR locus, though a potential influence of the lower number of SSRs loci used should not be discarded. When analysing null alleles presence and their effect on population structure, only 10 of the 406 loci harbored null alleles with a frequency higher than 0.20, and as such, no overestimation of F_ST due to null alleles was observed (mean FSTENA = 0.138 vs. mean F_ST = 0.136).

Overall, considering the above analysis, one can predict that the SSRs loci selected to screen the rye accessions under study are suitable for downstream genetic diversity analysis.

Rye samples analyzed displayed different levels of diversity (number of alleles and heterozygosity). Overall, cultivars displayed lower mean allele number (3) than landraces (4), but similar genetic diversity was observed as denoted by levels of observed heterozygosity (average H_OCultivars = 0.70, H_OLandraces = 0.67, Table 3). Accordingly, Matos et al. (2001) obtained a similar genetic diversity between Portuguese rye landraces and cultivars maintained in a Portuguese germplasm. Additionally, previous studies using RAPDs (Persson et al., 2001) and allozyme markers (Persson and von Bothmer, 2002, 2000) on Northern Europe rye showed that landraces and cultivars maintain roughly the same levels of genetic diversity. Overall, rye cultivars analyzed displayed similar genetic diversity as landraces, thus showing an unexpected absence of reduction in genetic diversity with increased improvement level, as recently reported for rye (Parat et al., 2016).

When narrowing analysis to landraces alone, ex situ collections revealed less allele per locus (3) than in situ (5) although showing similar heterozygosity (H_O ex situ = 0.67 vs. H_O in situ = 0.68, Table 3). This result is rather surprising, since generally higher genetic diversity would be expected within in situ compared to ex situ collections (e.g., Hou et al., 2012; Andrianasolo et al., 2013). In this study, ex situ and in situ accessions exhibit analogous genetic diversity levels, yet with different allelic diversity which may be indicative of a distinct subset of core alleles since ex situ are maintained under a steady environment without any selective pressure while in situ accessions are selected and grown by farmers in a subsistence agriculture context, being adapted under specific environments.

AMOVA showed that the majority of the genetic diversity lies within rye accessions with only little additional diversity present among groups or within groups. The large proportion of diversity found within accessions for the two types of groupings (cultivars vs. landraces; ex situ vs. in situ) suggests a high gene flow between accessions, which may be attributed to the wind-pollinated reproduction of rye allied with its outcrossing habit. One can therefore depict that rye-breeding system plays an important role in driving the generally observed high diversity within-accessions. Outcrossing plant species tend to have higher genetic variation within-populations, whereas selfing species or species with a mixed mating system are often genetically less variable (Nybom, 2004). Since rye is an outbreeder, negative to low inbreeding coefficients (F) were expected, which is in agreement with a recent rye study using SNPs (Hagenblad et al., 2016) but contrariwise to a SSRs study (Parat et al., 2016), where most populations displayed positive F-values which can be recorded in outcrossing populations with a strong population substructure. In the present study, low to moderate population differentiation among cultivars/ex situ was observed confirming the assumption of considerable population structure within cultivars and ex situ collection, which display steady population configuration as they are sequentially conserved in gene banks. Instead, in situ collections showed no population differentiation (low F_ST), which may be suggestive of high gene flow between in situ populations from Northeast Portugal and/or an evidence of a single recent genetic source.

Generally, wind-pollinated species require a great isolation distance, since its windborne pollen may travel reasonably large distances, while in insect-pollinated species distances are related to insect activity (Richards, 1986). The low correlation between genetic and geographic distances obtained, is in agreement with earlier analysis (Hagenblad et al., 2016), and can be explained considering that pollen transfer between cultivated rye fields may easily occur as a result of its wind-pollination mode thus requiring relatively high isolation distances as established for other cross-pollinated crops (e.g., 1000–1600 m for cabbage, cauliflower). This lack of correlation between genetic and geographic distances is also in accordance with previous assumptions that attribute both temporal and ecological isolation for shaping rye's genetic diversity (Ma et al., 2004) in deterrence to spatial or geographic isolation. Moreover, individual-based clustering methods (STRUCTURE and DAPC) applied to our data highlights that genetic diversity scattering does not follow a geographic trend, regardless of being cultivars or landraces. Previous studies support the lack of clear structuring of the distribution of genetic diversity in different rye accessions depicted from geography, by using classical (i.e., allozymes, Persson and von Bothmer, 2000, 2002; RAPDs, Persson et al., 2001; AFLPs, Chikmawati et al., 2005, 2012; SSRs, Akhavan et al., 2010; Targońska et al., 2016) and modern (DaRT, Bolibok-Brągoszewska et al., 2014) molecular marker systems as well as organellar genome diversity analysis (Isik et al., 2007). However, a recent study using SNPs revealed a clustering of European landraces according to its geographic origins (Hagenblad et al., 2016). Likewise, Parat et al. (2016) managed to obtain two main subgroups indicating a differentiation according to both geography and end use, which can be described as “southern European forage rye” vs. “northern European grain rye.” In our data, despite no structuring was depicted along a geographical array, STRUCTURE analysis revealed two main subgroups indicating a differentiation between cultivars/ex situ with in situ/rye's crop wild relative. Genetic structuring between rye cultivars and landraces has been reported earlier (Persson and von Bothmer, 2000; Bolibok-Brągoszewska et al., 2014; Targońska et al., 2016), evidencing a genetic diversity assortment according to breeding status. In our data, population structuring obtained highlights a similar allelic diversity between in situ collections and S. strictum, while cultivars and ex situ collections do not seem to share alleles with CWR. Likewise, a multivariate approach using DAPC show that cultivars exhibit shared allelic diversity with ex situ accessions while in situ accessions might retain allelic diversity similar with the rye CWR (K = 2) or not (K = 3). Population structuring of cultivars with ex situ collections is further reinforced by the UPGMA analysis, which also highlights a highly supported clade consisting of in situ collections with the rye CWR. Altogether, our data provides evidences of cultivars and ex situ collections harboring a different genetic diversity in contrast to in situ accessions. As ex situ accessions offers a static genetic snapshot, reflecting a population's adaptation to environmental conditions where they were collected, and considering that cultivars are a result of a controlled breeding process, a common gene pool can be depicted between cultivars and ex situ accessions here analyzed from eight European regions. Particularly, in situ populations screened in our study are originated from the region of excellence for rye production in Portugal, with diversity being preserved under farmer management. A recent study using dominant markers in three regional populations from the Northern Portugal shows the clustering of Portuguese populations in a different set than other rye cultivars, i.e., “Imperial,” “Dankowskie Zlote,” and the Portuguese cultivar “Alvão” (Santos et al., 2016), which is in accordance to our results. Considering the dissimilar population structuring of these Portuguese in situ populations with cultivars and ex situ collections, it cannot be ruled out that the region studied may be an hotspot of rye genetic diversity yet to be explored, and thus can provide valuable knowledge about genetic diversity resulting as part of the selection process adopted by local farmers through their agricultural practices.

Northeast Portuguese landraces have always been of great importance for local farmers, yet few studies have been performed for addressing its genetic diversity and population structure (Matos et al., 2001; Ribeiro et al., 2012; Santos et al., 2016). Adding other accessions with different geographic origins allows performing a comprehensive assessment of the genetic diversity of Portuguese landraces. Parat et al. (2016) by studying weedy, forage and grain ryes, which included a Portuguese forage landrace, determined a high fragmentation of membership coefficients in STRUCTURE analysis, which reflects a high diversity within Portuguese accessions. Indeed, our results support this former finding, with in situ Portuguese accessions displaying a high fragmentation with cultivars and ex situ collections. Interestingly, two ex situ accessions, “Sved” from Sweden and “Riodeva” from Spain, grouped along in situ accessions, as depicted by both UPGMA and model-based clustering (STRUCTURE and DAPC, K = 3) analysis. Interestingly, one cultivar (“Kungs II”) and one ex situ accession (“R780Spain”) originated from similar geographical regions as “Sved” and “Riodeva,” respectively, displayed a distinct genetic diversity from the former accessions and with the Portuguese in situ populations, thus highlighting the genetic distinctiveness of such rye accessions.

“Riodeva” is a Spanish rye inbred line resulting from a selection of a local landrace of Riodeva region (Lacadena et al., 1969), further bred over 30 generations of selfing (Gallego and Benito, 1997). In our study, admixture of “Riodeva” with Portuguese in situ populations was disclosed, in contrary with recent data using dominant markers (Santos et al., 2016), which showed no clustering of “Riodeva” with three regional Northern Portuguese populations. In our study, notwithstanding “Riodeva” displayed a lower mean number of alleles (Supplementary Table S3) comparing to the other rye accessions screened, a higher genetic diversity was obtained with the codominant markers used. “Riodeva” inbred-line has been used as a control for aluminum (Al) susceptibility (Gallego and Benito, 1997) and studies on rye aluminum tolerance have included this accession (e.g., De Sousa et al., 2016). Therefore, admixture between Portuguese in situ accessions with “Riodeva” can only be related to an ancient genetic diversity that remained even after the selfing process or, to some extent, to a potential relation with Al sensitivity/tolerance in acidic soils, since rye is one of the most tolerant cereals to Al-stress, with Portuguese in situ populations could hold different Al-tolerance behavior yet to be uncovered. Nevertheless, further genomic studies will be needed to disclose genetic background shared between Portuguese in situ populations and “Riodeva,” along with the characterization of Al-tolerance behavior.

“Sved” accession is a Swedish rye landrace that clustered with another accession (also from Finnmarken, on the border between Norway and Sweden) in a way distinct from all other Scandinavian and European rye landraces (Hagenblad et al., 2016), demonstrating to be a distinct genotype not found earlier in other rye landraces, including those from the same geographical provenance. Hagenblad et al. (2016) linked its distinctiveness with historical human migrations, since Finnish farmers settled after leaving their native country in the sixteenth century (Ahokas, 2008). Considering Portuguese historic trading markets it cannot be ruled out a scenario of multiple rye introductions into Portuguese territory, especially from Northern Europe (i.e., Sweden). Indeed, historical records evidence rye grain being conveyed from Sweden to Portugal in the late eighteenth century (Ojal and Karvonen, 2012). Thus, the observed genetic similarity of a Swedish landrace with Portuguese accessions could be ascribed to a historical context. Overall, in situ accessions displayed a genetic kinship with a distinct landrace genotype (“Sved”) along with rye CWR, highlighting a hidden diversity on Portuguese rye gene pool yet to be uncovered.

In cross-fertilized species like rye, open-pollinated varieties (OPVs) constitute panmictic populations harboring high levels of genetic variation in their genetic build-up (Geiger and Miedaner, 2009). Moreover, improved varieties are grown in relatively uniform agricultural environments, which tend to narrow its genetic pool. Therefore, a high phenotypic variation exhibited by improved varieties might not always be a good predictor of the extent of their genetic variation (McCouch, 2004), and to surpass this concern both in situ and ex situ approaches are used to conserve the genetic diversity (Gepts, 2006). Our study shows unequivocally that ex situ collections display a similar genetic architecture with cultivars, sharing genetic material in a great extent with “Petkus,” one of rye's heterotic pool (Hepting, 1978). Considering that not all parental lines could be uncovered from the cultivars used in our study, and that, as far as we know, none of the used cultivars have in its pedigree “Carsten” as a parental line, we can only infer about “Petkus” genetic pool. As such, ex situ collections screened do not present an effective alternative for supplementing “Petkus” pool. In contrast, in situ Portuguese collections displayed a significant different genetic diversity than cultivars, including “Petkus,” thus being surprisingly distinctive genotypes from both cultivars and ex situ collections. Thus, Portuguese rye gene pool will be important for identifying new useful alleles that are linked to local adaptive processes and to its ends use, either forage or grain. This is an important finding as it sheds light onto new rye genotypes that remain to be uncovered and that could be useful for incrementing “Petkus” genetic pool. A recent study in wheat genetic diversity uncovers a pool of regional divergence, and highlights the need to increase regional breeding programs for the maintenance of crop diversity, rather than consolidation of commercial breeding alone (Novoselović et al., 2016). It is unquestionable that conservation of agricultural in situ genetic resources provide the genetic building blocks to improve plant varieties, and our findings unfold new in situ resources that will boost the improvement of new rye varieties delivering innovative information to rye breeders.

Conservation of PGR through ex situ and in situ strategies have been implemented worldwide, yet in the last decades, there has been a growing interest in on farm conservation of landraces highlighted in the Convention of Biological Diversity (CBD), Agenda 21, and the International Treaty on Plant Genetic Resources for Food and Agriculture (ITPGRFA), emphasizing the importance of on farm conservation as an essential component of sustainable agriculture. By assessing in situ genetic diversity one can determine which landraces may hold “new” genetic variation that could be useful to supplement crop cultivars diversity, in traits with agronomical importance (i.e., abiotic and biotic stress). Such genetic diversity is generally concentrated in centers of diversity as well as on farm conditions since landraces structure and dynamics result from both natural and human selection (Gepts, 2006). Such rich agro-historical heritage requires conservation policies to preserve the management of landraces in farmers' fields where they originated, with the aim of maintaining the evolutive processes. The outcome of on farm conservation can be conceptualized as “evolutionary service” to agricultural and food systems, and to function, it depends on farmers' preferences, knowledge, management, practices, and social organization.

On farm conservation of local landraces, particularly those found in our study, reflects a specific case, since we studied landraces not listed at both national and international gene banks. As such, we believe that the first intervention toward a conservation protocol is to implement a new national seed policy on landraces, which usually favors only varieties that are distinct, uniform and stable, discouraging the use of more heterogeneous, variable landraces. In agricultural systems as the one practiced in Northern Portugal rye fields, farmers typically save seed from one season to the next and may share seed with other farmers, being seed sourcing embedded in well-structured traditional systems with rules and expectations based on family and local social networks (Veteläinen et al., 2009). As such, on farm conservation protocols should be compatible with improved livelihoods and well-being among farmers who conserve such landraces, by incrementing ecosystem services at regional and national level and by giving public benefits as a stimulus to promote specialized or novel marketing niches based on landraces and on local cultural heritage. Such landraces should be maintained as in situ genetic reserves and, besides ex situ conservation at national and international gene banks, an inventory periodically updated should be pursued to monitor on farm maintenance of landrace diversity.

Overall, our study successfully illustrates the significance of comparing the genetic diversity and structure of ex situ and in situ samples, along rye cultivars thus highlighting in situ collections from Northeast Portugal as new genetic resources being distinct genotypes to those reported for rye ex situ and cultivars. Identification of alleles/genes underlying such distinctive diversity would be of utmost importance for determining their usefulness for incorporating future rye breeding programs and to additionally propose on farm conservation policies at national level.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.