Introduction

The loss and fragmentation of natural and semi-natural habitats are the major causes of biodiversity decline in European agricultural landscapes (Liira et al., 2008). Fragmentation of habitats has decreased genetic diversity within populations and increased genetic differentiation among populations (Leimu et al., 2006) that, in turn, reduced their fitness and survival. Therefore, enhancing connectivity among fragmented habitats is one of the major goals of conservation planning (Brückmann et al., 2010).

A wide range of measures has been developed to quantify habitat connectivity that either measure structural connectivity accounting for the area and spatial configuration of habitats or quantify functional connectivity reflecting the actual movement of individuals or genes among populations (Tischendorf and Fahring, 2000). Structural connectivity can be relatively easily measured, for example, by calculating landscape metrics obtained from geographic information systems, assessing the topology of habitat or population networks or simply estimating geographic distances among habitats or populations. Therefore, structural connectivity is often used as a proxy for functional connectivity, although the response of species may not necessarily be correlated to the spatial configuration and composition of a landscape (Tischendorf and Fahring, 2000). However, success in restoration can only be achieved by evaluating to which degree an increase in structural connectivity (for example, the creation of stepping stones) improves the functional connectivity of study species. Functional connectivity can be measured by either directly observing the movement of species, for example, by radio-tracking or mark-recapture studies (Keller et al., 2010), or indirectly by using population genetic data (Keller et al., 2010). In plants, direct measurements of seed or pollen dispersal are very labour- and time consuming and may strongly underestimate the spatial and temporal extent of dispersal (Kamm et al., 2010). Therefore, genetic information has been successfully applied to assess functional connectivity in plant populations (Kamm et al., 2010).

One of the possibilities for examining the relationship between structural and functional connectivity of plants is to correlate measurements of gene flow to various measurements of landscape structure. Many studies simply correlated geographic distance with gene flow estimates among spatially isolated populations (see, for example Jacquemyn et al., 2007). However, in addition to geographic distance, landscape properties influence functional connectivity among populations by either fostering or impeding gene flow. Landscape genetic approaches have been widely applied to examine the relationship between landscape structure and functional connectivity in animal species, but, to date, only few studies applied landscape genetic approaches to examine gene flow among plant populations (Holderegger et al., 2010). This is most probably because of the fact that gene flow among plant populations is mediated through other vectors than plants themselves, for example, pollinators, wind, water and so on. Thus, it is not the plants that landscape structure acts upon, but seed and pollen dispersers, which complicates evaluating the possible role of various landscape characteristics on gene flow in plants.

Graph-theoretic connectivity metrics combine the attributes of habitats and populations with the dispersal behaviour of species (Pascual-Hortal and Saura, 2006). In graph-theoretic approaches, the habitat patches or populations (nodes) are connected by links, for example, geographic distances (Saura and Pascual-Hortal, 2007; Saura and Torne, 2009). Nodes are considered connected if the internode distance is lower than, for example, the maximum dispersal distance of a focal species (that is, structural connectivity). Based on the topology of a network, it is possible to determine the most critical network elements—nodes as well as links—for maintaining the structural connectivity of habitat patches or populations in a landscape. This approach has been suggested as a useful tool in assisting decision making for conservation purposes (Pascual-Hortal and Saura, 2006). Despite the increasing application of graph-based connectivity metrics in recent years, studies relating these measures to estimates of functional connectivity, such as gene flow, or to genetic diversity, remain scarce (Neel, 2008; Dyer et al., 2012).

The current study was carried out in an intensively managed Swiss agricultural landscape where semi-natural grasslands made up only 2% of the total area. We examined whether these fragmented grasslands within a matrix of agricultural land, forests and settlements maintain gene flow and genetic diversity as measured from molecular markers from Lychnis flos-cuculi, an insect-pollinated and predominantly outcrossing wetland plant with no specialised seed dispersal mechanism. This plant species was chosen because insect-pollinated outcrossing species are most vulnerable to habitat fragmentation (Aguilar et al., 2008). First, we evaluated whether graph-based estimates of structural connectivity reflect functional connectivity, measured as gene flow, and genetic diversity of this species. Second, we examined the effects of landscape elements on gene flow in L. flos-cuculi using a landscape genetic approach based on transects (Holderegger et al., 2010; Angelone et al., 2011; Van Strien et al., 2012). We hypothesised that settlements, agricultural land, forests and topography (that is, elevation change) negatively affect gene flow (Tewksbury et al., 2002; Helm et al., 2009; Keller et al., 2012; Pellissier et al., 2012), whereas the amount of ditch verges being one of the key habitats of this species in contemporary agricultural landscapes (Aavik et al., 2008) facilitate gene flow in L. flos-cuculi.

Materials and methods

Study species, data collection and genetic analyses

L. flos-cuculi L. (syn. Silene flos-cuculi; Caryophyllaceae) is a diploid polycarpic perennial herb that is distributed throughout Europe (Jalas and Suominen, 1986). It usually grows in moist, open habitats such as floodplain and fen meadows. In arable landscapes, it has become scarce because of loss of habitats, but is still found in secondary habitats such as ditch and stream verges. L. flos-cuculi is an insect-pollinated species visited by various pollinators: Lepidoptera, Diptera and Hymenoptera (Van Rossum and Triest, 2010). The species is predominantly outcrossing, but capable of self-fertilisation (Biere, 1991). Stems are 30–90 cm tall, and flowers open between April and June in the second year after establishing. Capsules contain ∼150 seeds (Biere, 1991) that are dispersed mechanically. In addition to sexual reproduction, L. flos-cuculi forms vegetative rosettes from axillary stem buds.

The study area with a size of ∼80 km2 is an agricultural landscape with a mixture of pastures and arable land (Figure 1). It is located in the Cantons of Bern and Aargau in Switzerland. We mapped the centroids of all naturally occurring populations of L. flos-cuculi in the study area using Global Positioning System, collected leaves from 30 individuals (where possible) of L. flos-cuculi in each population and estimated census population size (Table 1).

Figure 1
figure 1

Location of study populations of Lychnis flos-cuculi in Switzerland. Grey lines mark ditches and other water bodies with flowing water; grey areas indicate forested areas.

Full size image
Table 1 Coordinates (according to Swiss national grid), sample and population size, allelic richness (AR), expected heterozygosity (HE), observed heterozygosity (HO) and inbreeding coefficient (FIS) of the study populations of Lychnis flos-cuculi
Full size table

We kept collected plant material in silica gel until DNA extraction. DNA was extracted from 10 mg of dry leaf material using the Dneasy 96 Plant Kit (QIAGEN, Hombrechtikon, Switzerland). We used three microsatellite markers developed for L. flos-cuculi (Galeuchet et al., 2002) and three markers developed for Silene latifolia (Moccia et al., 2009), a close relative of L. flos-cuculi. PCR was carried out as described in Aavik et al. (2012). PCR products were analysed on an ABI 3730 automated sequencer (Applied Biosystems, Carlsbad, CA, USA). Allele lengths were scored using GENEMAPPER 3.7 (Applied Biosystems). Tests of linkage disequilibrium in FSTAT 2.9.3.2 (Goudet, 1995) over all populations revealed a significant linkage between three pairs of loci, but a lack of significant linkage within populations indicated that those loci were not physically linked. FSTAT 2.9.3.2 was used for detecting deviations of loci from Hardy–Weinberg expectations, but none of the loci deviated from Hardy–Weinberg expectations. As estimates of genetic diversity within populations, we used allelic richness (AR, indicating the number of alleles per locus based on a minimum sample size of 13 individuals), expected heterozygosity (HE), observed heterozygosity (HO) and inbreeding coefficient (FIS), which were all calculated using FSTAT 2.9.3.2 (Goudet, 1995).

Landscape data

Transect- or corridor-based analysis is often used in landscape genetics of plants (Holderegger et al., 2010). We therefore generated pairwise straight lines between all populations of L. flos-cuculi in ARCGIS 9.3.1 (ESRI, Redlands, CA, USA). The lines were buffered to create corridors with different widths of 50, 100, 300, 500 and 1000 m. Within these corridors, the proportion of several landscape elements potentially influencing gene flow by seed or pollen of L. flos-cuculi was determined based on VECTOR25 data (Swisstopo, Switzerland; http://www.swisstopo.admin.ch/internet/swisstopo/en/home/products/landscape/vector25.html). VECTOR25 provides nine thematic layers among which the layer of primary surfaces was used in the present study. We calculated the proportion of the following landscape elements that can potentially hamper gene flow of plants: settlements (Helm et al., 2009), agricultural land and forests (Tewksbury et al., 2002). In addition, we calculated the area along the verges of ditches as suitable habitats of L. flos-cuculi in fragmented agricultural landscapes (Aavik et al., 2008) by buffering ditches by 2 m on both sides. We also calculated Euclidean distance using ARCGIS 9.3 and the ‘ArcGis toolbox’ of landscape genetics (Etherington, 2011), as well as the cumulative elevation change between all populations using a digital height model with a 25 m resolution (DHM25; Swisstopo, Switzerland), because strong elevation gradients can significantly hinder the movement of insects (Keller et al., 2012) as potential pollen vectors of L. flos-cuculi.

Network analysis

We used CONEFOR 2.6 (Pascual-Hortal and Saura, 2006; Saura and Pascual-Hortal, 2007; Saura and Torne, 2009) to assess the relative importance of populations (nodes) and links (geographic distance) in the population network of L. flos-cuculi. CONEFOR 2.6 assesses a node or a link by comparing the distance between nodes and a species-specific threshold dispersal distance for the investigated organism. If the distance between two nodes is shorter than the assigned threshold dispersal distance, nodes are considered as equally connected (Pascual-Hortal and Saura, 2006; Saura and Pascual-Hortal, 2007; Saura and Torne, 2009). Previous observations about dispersal and pollination distances of L. flos-cuculi did not come up with a clear threshold distance for this species (Van Rossum and Triest, 2010; Aavik et al., 2013). Therefore, we tested various threshold distances beyond which all nodes were considered equally connected (that is, 100, 200, 500, 1000, 2000, 3000 and 5000 m). Nodes were characterised by log-transformed population sizes, and links were evaluated using Euclidean distances between populations.

Two graph-based connectivity indices, the integral index of connectivity (IIC) and the probability of connectivity (PC), have been shown to perform best for the purpose of prioritisation of habitats and links (Pascual-Hortal and Saura, 2006; Saura and Torne, 2009). We therefore chose these two indices for evaluating the importance of nodes and links in the present study. IIC ranges between 0 and 1 and is defined as:

where n is the total number of nodes in the landscape, ai, and aj are the attributes of nodes i and j and nlij is the number of links (distances) between nodes i and j. IIC is based on a binary connection model that means that two patches considered are either connected or not depending on the actual distance between them in relation to the predefined distance threshold.

Similar to IIC, PC ranges between 0 and 1 (Saura and Pascual-Hortal, 2007) and is calculated as:

where pij* is defined as the maximum product probability of all possible paths between nodes i and j. In contrast to IIC, PC is based on a probabilistic connection model, where a certain probability of dispersal between the two patches considered (pij) characterises the links between nodes i and j in the graph. Because of lack of adequate background information about the potential dispersal distance of L. flos-cuculi, a probability of 0.5 was used for connectivity calculations that corresponds to the median dispersal distance of the study species depending on the threshold distance considered.

In CONEFOR 2.6, the prioritisation of each node for maintaining landscape connectivity is calculated as the percentage of the variation of IIC or PC resulting from the removal of a given node from the landscape:

where I is the overall index value (both for IIC and PC) for all initially existing nodes in the landscape, and Iremove is the overall index value after the removal of a given node from the landscape.

The index includes the connections between nodes (interpopulation connectivity in the current study) as well as the connected area existing within the nodes themselves (intrapopulation connectivity; Saura and Pascual-Hortal, 2007). Saura and Rubio (2010) proposed a way for partitioning node importance into three fractions that quantify the way each node contributes to overall habitat connectivity. Accordingly, node importance consists of three components, depending on whether IIC or PC are considered:

and

where dIICintrak and dPCintrak represent the contribution of node k to intrapopulation connectivity, for example, habitat area or population size. This value is independent of the dispersal distance of the focal species and does not depend on the connectivity of the node to other nodes. In the present study, this index was calculated on the basis of the node property population size. dIICfluxk and dPCfluxk indicate how well a node k is connected to other nodes in the landscape (that is, the interpopulation connectivity of a node) without considering its contribution to intrapopulation connectivity. dIICconnectork and dPCconnectork show whether a node k contributes to the connectivity between other nodes as a stepping stone (Saura and Rubio, 2010). We calculated the three fractions of connectivity for each study population as described in Equations (4) and (5). However, dIICconnectork and dPCconnectork were mostly very close to or equal to zero and were therefore not considered in further analysis.

In addition to assessing the importance of each node, CONEFOR 2.6 enables evaluating the contribution of each individual link to landscape connectivity. For calculating link importance, we selected the ‘link removal’ option that assesses the impact of removing each of the links for landscape connectivity. The impact of losing a link from a network was calculated using both the binary connection model (IIC) and the probabilistic connection model (PC).

Statistical analysis

We used nonparametric γ-correlations to assess the relationship between estimates of genetic diversity (AR, HE, HO and FIS) and node importance (dIIC and dPC) as well as intra- and interpopulation components of node importance (dPCintra, dPCflux, dIICintra and dIICflux) with different threshold distances (100, 200, 500, 1000, 2000, 3000 and 5000 m). The γ-correlations were used because of tied data in the data set.

To estimate the effects of landscape variables on gene flow as a measure of functional connectivity, we used Euclidian distance, cumulative elevation change and the landscape variables agricultural land, settlements, forests and ditch verges calculated within corridors of different widths (50, 100, 300, 500 and 1000 m). First, we aimed at determining the most representative corridor width that best explained gene flow. To avoid multicollinearity, all landscape variables were checked for Spearman’s rank correlation. All correlations between landscape variables were below rs<0.6 with the exception of Euclidian distance and cumulative elevation change (rs>0.8). Therefore, the latter was excluded from further analysis. Pairwise genetic differentiation FST (Wright, 1965), a measure of past gene flow among populations, was calculated in FSTAT 2.9.3.2 (Goudet, 1995) and used as a response variable. The influence of Euclidian distance, agricultural land, forest, settlements and ditch verges on FST was then evaluated with multiple regression on distance matrices using function ‘MRM’ (package ecodist; Goslee and Urban, 2007) in R 2.15.1 (R Development Core Team, 2012). All landscape variables were rank-transformed before entering regression analysis, and the significance of regression coefficients was estimated using 10 000 permutations. Separate regressions were performed on models with different corridor widths. For further analysis, we used the landscape variables calculated within corridors of 300 m width, because they provided the best goodness of fit (Table 2).

Table 2 Results of regression on distance matrices analyses on the effects of landscape variables on the genetic differentiation FST among the study populations of Lychnis flos-cuculi in Switzerland using different corridor widths
Full size table

In a next step we built regression models on distance matrices that included pairwise FST as a response variable and the following structural connectivity variables as explanatory parameters: link importance, pairwise summarised values of intra- and interpopulation connectivity and landscape variables calculated within corridors of 300 m width. Separate sets of models were built for different threshold distances (100, 200, 500, 1000, 2000, 3000 and 5000 m) for both network measures, PC and IIC. Explanatory variables exhibiting a correlation of rs⩾0.7 (Dormann et al., 2013) were excluded from analysis to avoid multicollinearity. Some variables could not be included in the analysis because of missing values (that is, summarised value of interpopulation connectivity and link importance at distance thresholds of 100 and 200 m in IIC). Explanatory variables were again rank-transformed before performing the analysis. The significance of regression coefficients was estimated using 10 000 permutations.

Results

The effect of structural graph-theoretic connectivity measures on genetic diversity

Depending on the distance threshold and connectivity index considered, we detected significant correlations between node importance and estimates of genetic diversity (Table 3). Node importance based on the binary connection model (dIIC) was significantly and positively related to observed (HO) and expected heterozygosity (HE) at the distance threshold of 500 m and negatively related to the inbreeding coefficient (FIS) at a threshold of 1000 m (Table 3 and Figure 2a). Node importance based on the probability of connectivity (dPC) had a positive relationship with HO at distance thresholds of 200 and 500 m (Table 3 and Figure 2b). The interpopulation connectivity component of node importance based on IIC (dIICflux) was positively related to HO at distance thresholds of 500–2000 m, but had a negative relationship with FIS at thresholds of 500 and 1000 m (Table 3 and Figure 2c). The interpopulation component of node importance based on PC (dPCflux) had a positive relationship with HO at thresholds of 100–1000 m and a negative correlation with FIS at 100–500 m (Table 3 and Figure 2d). In conclusion, correlations between estimates of genetic diversity and node importance as well as between genetic diversity and the interpopulation connectivity component of node importance were highest at distances ranging between 500 and 1000 m. Correlations were lowest at larger distance thresholds. The intrapopulation connectivity component (dIICintra and dPCintra) was not related to any of the variables of genetic diversity.

Table 3 The γ-correlations between genetic diversity (HE: expected heterozygosity; HO: observed heterozygosity; AR: allelic richness; FIS: inbreeding coefficients) and node importance (dIIC: integral index of connectivity; dPC: probability of connectivity) and between genetic diversity and different components of node importance (dIICintra, dIICflux, dPCintra, dPCflux) with distance thresholds ranging from 100 to 5000 m
Full size table
Figure 2
figure 2

Correlation coefficients between estimates of genetic diversity within the populations of Lychnis flos-cuculi and node importance (a: integral index of connectivity (dIIC); b: probability connectivity index (dPC)) and between estimates of genetic diversity and interpopulation connectivity components of node importance (c: metric based on integral index of connectivity (dIICflux); d: metric based on probability connectivity index (dPCflux)) using different threshold distances (100–5000 m).

Full size image

The effect of landscape and graph-theoretic connectivity measures on gene flow

Multiple regression on distance matrices revealed that among landscape variables, forest had a significantly positive effect on genetic differentiation FST (that is, a negative effect on gene flow) in all models (Table 4). For the remaining landscape variables, only the proportion of ditch verges had a minor negative effect on gene flow, generally in those models that included connectivity metrics calculated at larger distance thresholds (⩾3000 m). Among variables of structural graph-based connectivity, the summarised value of intrapopulation component of node importance (dIICintra or dPCintra) had a significantly negative effect on genetic differentiation (that is, a positive effect on gene flow) in most of the models, particularly in models including connectivity metrics calculated at larger distance thresholds (⩾1000 m; Table 4). The summarised pairwise value of interpopulation component (dIICflux and dPCflux) had almost no effect on genetic differentiation. Similarly, link importance had no effect on gene flow. Surprisingly, geographic distance remained nonsignificant in all models. The goodness of fit (R2) of models (all including the same number of explanatory variables) ranged between 0.155 and 0.332, explaining a reasonable part of the variation in genetic differentiation. Goodness of fit was relatively low (R2<0.2) in models including connectivity metrics calculated at lower distance thresholds and started to increase in models including connectivity metrics at distance thresholds beyond 1000 m.

Table 4 Results of multiple regression on distance matrices on the effect of landscape and graph-theoretic connectivity variables on gene flow of Lychnis flos-cuculi using different threshold distances (100 –5000 m) and connectivity indices (IIC and PC)
Full size table

Discussion

Maintaining and enhancing spatial connectivity among fragmented natural and semi-natural habitats is one of the central targets of nature conservation. However, though the structural properties of landscapes are widely used as indicators of biodiversity in practical landscape planning, they may not have a straightforward effect on functional connectivity, that is, the way individual species respond to landscape structure (Tischendorf and Fahring, 2000). In the present study, we used the proportion of several landscape variables and two graph-theory based indices (Pascual-Hortal and Saura, 2006)—IIC and PC—to quantify structural connectivity among the study populations of L. flos-cuculi occurring in fragmented grassland patches of an intensively managed agricultural landscape. In addition, gene flow among populations was used as a measure of functional connectivity. This enabled us to examine the relationship between structural and functional connectivity of this formerly common grassland species, whose abundance has recently decreased because of landscape fragmentation and habitat loss.

Relationship between structural connectivity and genetic diversity

Genetic diversity measured by expected and observed heterozygosity within populations of L. flos-cuculi was positively related to node importance (that is, the contribution of populations to structural connectivity) at distance thresholds between 200 and 1000 m. Similarly, the interpopulation connectivity fraction of node importance calculated at distance thresholds of 100–1000 m was more strongly related to indicators of genetic diversity than to connectivity parameters with thresholds >1000 m. These results probably indicate that gene flow in L. flos-cuculi takes place at distances of <1 km. Gene flow of this species can occur by both insect-mediated pollen flow and seed dispersal. Generally, most of the pollen flow of insect-pollinated species is restricted to the nearest vicinity of plants (Van Rossum and Triest, 2010), although occasional pollen flow events occur over much longer distances (Kamm et al., 2010). Likewise, seeds of L. flos-cuculi seldom disperse far from source populations (Aavik et al., 2013): most seeds are spread in the close neighbourhood of mother plants because of the lack of a specialised dispersal syndrome. Thus, geographic distance between isolated populations of grassland plants with seed and pollen dispersal strategies similar to L. flos-cuculi should not be larger than ∼1 km in order to maintain genetic diversity within populations.

We did not detect any significant relationship between genetic diversity and intrapopulation connectivity component. In the present study, this index was based on population size, and Leimu et al. (2006) showed that larger populations are generally characterised by higher genetic diversity. However, small populations may still contain substantial genetic variation if they represent remnants of recently larger populations (Landergott et al., 2001; Klank et al., 2012) and thus the relationship between population size or related indices and genetic diversity may not have re-established yet.

Indices based on PC have shown to perform better as compared with other graph-based connectivity measures (Saura and Pascual-Hortal, 2007), whereas the IIC may oversimplify connectivity among populations because of its binary approach. Neel (2008), in contrast, found that connectivity measures related to IIC predicted estimates of genetic diversity within plant populations better than those based on PC. In the current study, we did not detect any notable differences between these two indices.

The effects of landscape elements on gene flow

Among the studied landscape elements, only forest had a consistent and significantly negative effect on gene flow among populations of L. flos-cuculi. Previous studies demonstrated that not only patches of forest (Schmitt et al., 2000), but also linear woody elements considerably restrict the movement of certain pollinator groups in an open landscape (Wratten et al., 2003). Furthermore, the presence of open corridors within a forest matrix significantly enhanced pollen flow among otherwise spatially isolated insect-pollinated plant populations (Tewksbury et al., 2002). Seeds of L. flos-cuculi do not have specialised dispersal mechanism and most of the seeds disperse over short distances. Thus, it is very likely that the negative effect of forest on gene flow among the study populations was because of forest inhibiting the movement of pollinators (but see Zurbuchen et al., 2010) rather than the dispersal of seeds.

We hypothesised that agricultural fields and settlements would hinder gene flow among the study populations (Helm et al., 2009; Pellissier et al., 2012), because agricultural intensification and urbanisation have been the major processes contributing to habitat fragmentation. Surprisingly, genetic differentiation was independent of both of these variables. One possible explanation is that the changes in the spatial structure of landscapes are not yet reflected in genetic differentiation FST among populations, the estimate for functional connectivity of L. flos-cuculi used in the present study, because perennial plant species respond to landscape fragmentation more slowly than short-lived species (Kuussaari et al., 2009). In addition, genetic differentiation does not necessarily reflect contemporary gene flow (Whitlock and McCauley, 1999). A previous study examining current levels of gene flow among the same populations of L. flos-cuculi as studied here revealed that present-day functional connectivity of this species is indeed low (Aavik et al., 2013), and therefore it is likely that genetic differentiation FST rather represents historical than current functional connectivity in our study landscape. The most dramatic landscape changes in Switzerland took place in the second half of the twentieth century (Schultz and Dosch, 2005), and thus the effects of these changes on the genetic differentiation of perennial plants might only become traceable in the future (Honnay et al., 2006; Klank et al., 2012).

It is also possible that agricultural land does indeed not inhibit gene flow. Recent studies have demonstrated that open areas in European agricultural landscapes do in fact enhance pollen flow among spatially isolated insect-pollinated plants (Kamm et al., 2010). Furthermore, agricultural areas with a network of structurally connected semi-natural linear elements still contain small stepping-stone plant populations that support higher levels of gene flow (Van Geert et al., 2010; Van Rossum and Triest, 2012). In our study area, field boundaries adjacent to roads or forests were the most common semi-natural habitats and could have mitigated the potential negative effect of agricultural fields on gene flow in L. flos-cuculi, at least to a certain extent. However, note that a higher proportion of ditch verges among our study populations had no or even a slightly negative effect on gene flow according to most models (Table 4).

The relationship between graph-theoretic connectivity measures and gene flow

We found that the intrapopulation fraction of structural connectivity had a significant positive effect on gene flow, that is, a pair of populations having a higher summarised value of intrapopulation connectivity was characterised by lower genetic differentiation. In the present study, we calculated the contribution of a population to intrapopulation connectivity based on its population size. Our findings thus indicate that more gene flow is taking place among larger populations than among smaller ones. Several mechanisms related to either seed dispersal or pollen flow may explain this observed pattern. First, larger patches of flowering plants generally enhance gene flow by pollen (Dauber et al., 2010), as larger populations are visited by pollinators more frequently and thus serve as more important sources as well as sinks of pollen than smaller populations (Richards et al., 1999). Second, larger populations usually produce more dispersal propagules that have therefore a higher likelihood of dispersing into the surrounding landscape than the lower number of seeds from smaller populations (Ellstrand and Elam, 1993). It has also been shown that seeds from larger populations exhibit higher germination success and thus fitness than seeds from smaller populations (Faast et al., 2011), perhaps as a result of the higher genetic diversity observed in larger populations (Leimu et al., 2006). None of the other connectivity metrics (link importance and summarised value of interpopulation connectivity) had any effect on gene flow in the present study, regardless of the different threshold distances or connectivity indices (IIC and PC) considered. One of the reasons for the lack of a relationship between gene flow and link importance could be the above-mentioned delayed effect of structural changes on functional connectivity (Kuussaari et al., 2009).

Conclusions

We conclude that structural connectivity, as measured by graph-based approaches, influences genetic diversity within populations, but does not (or only in an indirect way) affect gene flow among populations of L. flos-cuculi. This suggests that structural connectivity should be used as a proxy for functional connectivity with caution (Tischendorf and Fahring, 2000). Our findings nevertheless show that a distinction between IIC and PC into fractions of intra- and inter-population connectivity could provide valuable insights into processes driving functional connectivity (Saura and Rubio, 2010).

Among the studied landscape variables, only forest had a negative effect on gene flow, most likely through hindering the movement of pollinators. Surprisingly, we did not detect any significant effect of Euclidian distance or of any other landscape variable on gene flow. Although it is possible that the negative effects of settlements and agricultural fields were partly mitigated by linear semi-natural elements connecting the spatially isolated populations of L. flos-cuculi, it is also possible that the effect of landscape fragmentation on genetic differentiation is subject to time lags, particularly in case of perennial species (Kuussaari et al., 2009). Thus, studies focussing on gene flow patterns of plants in recently fragmented landscapes may benefit from examining contemporary gene flow patterns.

Data archiving

Data available from the Dryad Digital Repository: doi:10.5061/dryad.147r0.