Patterns and processes in a non‐adaptive radiation: Alopia (Gastropoda, Clausiliidae) in the Bucegi Mountains

We studied the door‐snail genus Alopia in the Southern Carpathians in Romania to better understand non‐adaptive radiations and the processes that determine their course. Alopia in the Bucegi Mountains offers the opportunity to study all stages of a radiation within a few kilometres. The species (as defined by the differential fitness species concept) in the most advanced stage of the radiation differs from other species in the genitalia and co‐occurs with other species. The least advanced stages are genetically differentiated clusters of populations that are geographically separated from other clusters but are not yet morphologically differentiated. Differentiation does not increase with a constant rate as shown by a lineage that was separated early in the evolution of the genus but fused with other taxa upon secondary contact. Since non‐adaptive radiation cannot be accelerated by divergent natural selection beyond the slow pace of speciation that is possible by genetic drift alone, sexual selection plays a crucial role in non‐adaptive radiations. This is supported by the differences in the genitalia found in the most advanced stage of the speciation which indicate that the speciation process was accelerated by a co‐evolutionary arms race resulting in the elongation of spermatophore‐producing and spermatophore‐receiving organs. Another process that facilitates non‐adaptive radiation is long‐distance dispersal that results in geographically isolated populations that can differentiate without gene flow and that have a higher likelihood of speciation due to the founder event. Several taxa that were considered distinct species until now, but fuse upon contact should better be classified as subspecies of a polytypic species.


| 281
KOCH et al. result of ecological speciation by divergent selection and a rapid adaptation of incipient species to different niches, have been studied in detail, for example the Darwin's finches in the Galápagos Islands (Grant & Grant, 2008), Anolis lizards of the Caribbean (Losos, 2009) or cichlid fish in Africa (Seehausen, 2006). In contrast, non-adaptive radiations, in which diversification is not accompanied by adaptation into different niches, but results in a group of usually allopatric species occupying similar niches (Dominey, 1984;Fehér et al., 2018;Gittenberger, 1991;Rundell & Price, 2009) have been neglected for a long time. However, the importance of non-adaptive radiations may be underestimated. It is likely that in some apparently adaptive radiations non-adaptive speciation triggered by geographical isolation preceded the later ecological differentiation of the species (Rundell & Price, 2009). Thus, a better insight into non-adaptive radiations is essential for our understanding of the origin of biodiversity.
A radiation can be defined as the evolution of a relatively large monophyletic group of species (Gittenberger, 1991). Thus, we also have to define what a species is. We do not use the restrictive biological species concept (Mayr, 1942(Mayr, , 1963) that requires reproductive isolation. Many species in recent radiations like the ground finches (Grant & Grant, 2008;Lamichhaney et al., 2015) do not fulfil this criterion and, thus, would not be considered distinct species under this concept. Instead, we use the differential fitness species concept (Hausdorf, 2011) that defines species as groups of individuals that are reciprocally characterized by features that would have negative fitness effects in other groups and that cannot be regularly exchanged between groups upon contact. In an adaptive radiation, niche divergence is connected to speciation. Thus, the features that would have negative fitness effects in other groups may be adaptations to different environments or different functions in an environment. In a non-adaptive radiation, the niches of the emerging species do not differ distinctly. Thus, the speciation process is not driven by niche-related factors and depends only on features affecting reproduction or the intrinsic fitness of the offspring.
Non-adaptive radiations are frequent in groups with limited dispersal capability. The classical example of a non-adaptive radiation is a radiation of the door-snail (Clausiliidae) genus Albinaria inhabiting limestone rocks on Crete (Gittenberger, 1991). Many other snail taxa are restricted to limestone rocks, where they feed mainly on lichens. Limestone rock outcrops are often isolated habitat islands that are separated by areas without rocks. The geographical isolation of limestone outcrops facilitates the differentiation and radiation of rock-dwelling snail species. For example, there are several species-rich radiations of door-snail genera in southern Europe (Nordsieck, 2007).
In this study, we investigate the door-snail genus Alopia in the Bucegi Mountains in the southern Carpathians in Romania. Although this mountain massif covers less than 10 km × 20 km, up to ten endemic Alopia (sub-)species were recorded from that area (Fehér, Németh, Nicoară, & Szekeres, 2013;Grossu, 1981;Nordsieck, 2008). The mitochondrial gene tree of Alopia (Fehér et al., 2013) showed that the taxa inhabiting the Bucegi Mountains belong to three different clades, each of which has further representatives in other parts of the Carpathian Mountains and each of which is more closely related to clades from other parts of the Carpathian Mountains than to the geographically adjacent clades in the Bucegi Mountains. This can be explained by two hypotheses. Either the Bucegi Mountains were colonized three times by different Alopia clades or the Bucegi Mountains were part of the ancestral area of Alopia, where the different clades evolved and from where other regions of the Carpathian Mountains were colonized. The latter hypothesis had already been proposed by Kimakowicz (1894). In any case, the Alopia taxa from the Bucegi Mountains do not represent a separate radiation but are only a part of the Alopia radiation in the Carpathian Mountains. We focus here on the Bucegi Mountains because several taxa occur there in close proximity so that we can study potential gene flow between taxa, whereas in many other areas, different taxa are separated by wide geographical barriers so that migration between ranges of different taxa is very low.
We investigated the stages of the non-adaptive radiation of the Alopia taxa in the Bucegi Mountains and the processes that drive or counteract the differentiation process using mitochondrial DNA sequences as well as nuclear AFLP markers. We also propose a new classification of the Alopia taxa from the Bucegi Mountains considering their genetic differentiation and the admixture between population groups.

| Sampling and classification
Alopia populations from 24 locations ( Figure S1) were sampled in the Bucegi Mountains in the southern Carpathian Mountains in Romania in June 2013. Herilla ziegleri dacica, and Albinaria p. puella were used as outgroups for the phylogenetic analysis of cox1 sequences. The classification, locality data and voucher numbers of the specimens used in this study are compiled in Table S1. The samples were stored in 100% isopropanol at −20°C.
The samples were identified morphologically to the species and subspecies distinguished by Nordsieck (2008) and Fehér et al. (2013) (Table 1). In the Results section, we used the nomenclature proposed by Nordsieck (2008) with exception of the subspecies of A. livida (in the traditional sense). Both, Nordsieck (2008Nordsieck ( , 2015Nordsieck ( , 2016 and Fehér et al. (2013) separated the populations of A. livida from higher altitudes as a separate subspecies A. livida bipalatalis (Kimakowicz, 1883). Nordsieck (2008) replaced this name by A. livida sororcula Soós, 1928 because it is preoccupied by Clausilia bipalatalis Martens (in Boettger, 1878), but Nordsieck (2016) argued that A. livida sororcula is a synonym of Alopia livida hypula Soós, 1928 and used A. livida bipalatalis again. Alopia livida bipalatalis is allegedly characterized by the frequent possession of palatal folds. However, the percentage of specimens with palatal folds varies between 55% and 100% in the populations at higher altitudes and in populations at lower altitudes between Ciubotea and Grohotiş up to 35% of the specimens also possess palatal folds (Nordsieck, 2015). A. livida cannot be classified into subspecies based on morphological characters because of the lack of consistent differences between population groups. In the Results section, we preliminarily subdivide A. livida according to the network based on nuclear AFLP marker. We assigned available names to these population clusters based on their proximity to the type localities compiled in Table S2. We provisionally follow Nordsieck (2015Nordsieck ( , 2016 and Fehér et al. (2013) in using the name A. livida bipalatalis for the populations of A. livida from higher altitudes. Nordsieck (2015) argued that the type of A. livida was from the north-western foothills of the Bucegi Mountains between Ciubotea and Grohotiş, but included also the populations from the southern and south-eastern parts of the Bucegi Mountains in A. l. livida. However, these populations probably do not form a coherent evolutionary unit with those from the north-western foothills, because they are separated by populations at higher altitude along the ridge of the mountains that Nordsieck (2008Nordsieck ( , 2015

| DNA extraction, amplification and sequencing
Total genomic DNA was extracted from tissue samples of the foot following the protocol proposed by Sokolov (2000) with slight modifications as detailed in Scheel and Hausdorf (2012).

DNA sequences
Forward and reverse sequence reads were assembled using CHROMASPRO version 1.7.1 (Technelysium). The sequences were aligned with MUSCLE (Edgar, 2004) as implemented in MEGA version 7 (Kumar, Stecher, & Tamura, 2016) using the default settings. MEGA 7 was also used to calculate p-distances.
The cox1 alignment was divided into three partitions corresponding to the three codon positions. The HKY + G model was selected for all three partitions using the Akaike information criterion (AIC) implemented in Treefinder (Jobb, 2011;Jobb, von Haeseler, & Strimmer, 2004). Heuristic maximum-likelihood analyses were performed in Treefinder setting the search depth to 2 and allowing for independent estimation of parameters for individual partitions. Confidence values were computed by bootstrapping (1,000 replications).
Electropherograms were analysed with PEAK SCANNER version 1.0 (Applied Biosystems) to detect AFLP bands and calculate their size using default settings except for a light peak smoothing, sizing quality with a pass range of 0.1-1 and a low-quality range of 0.0. Fluorescent threshold was set to 50 relative fluorescence units. Binning and scoring were performed using RAWGENO version 2.0 (Arrigo, Tuszynski, Ehrich, Gerdes, & Alvarez, 2009), an add-on package for the statistical software suite R (R Core Team, 2012), with the following settings: scoring range = 50-500 bp, minimum intensity = 100 rfu, minimum bin width = 1.5 bp and maximum bin width = 2.0 bp. Bins with reproducibility lower than 80% (the default value of the reproducibility filter) were eliminated. A replicate reproducibility rate of 91.4% (n = 13) was calculated as the mean percentage of matching character states between replicates of the same individual (Pompanon, Bonin, Bellemain, & Taberlet, 2005). A neighbour-net (Bryant & Moulton, 2004) was constructed based on Jaccard distances calculated with the AFLP data using SPLITSTREE4 version 4.14.3 (Huson & Bryant, 2006).
Individual-based clustering and admixture analyses of the AFLP data were performed with STRUCTURE version 2.3.4 (Falush, Stephens, & Pritchard, 2007;Pritchard, Stephens, & Donnelly, 2000) as well as BAPS 6.0 (Corander & Marttinen, 2006;Corander, Marttinen, Sirén, & Tang, 2008). We carried out ten runs with 800,000 iterations after a burn-in of 200,000 iterations for each cluster number K from 1 to 12 with STRUCTURE. We used the mean estimates of the posterior probabilities of the data for a given cluster number L(K) and the ad hoc quantity ΔK proposed by Evanno, Regnaut, and Goudet, (2005) computed with STRUCTURE HARVESTER (Earl & vonHoldt, 2012) to estimate the number of clusters. DISTRUCT version 1.1 (Rosenberg, 2004) was used for visualizing the admixture calculated in the STRUCTURE run with the highest posterior probability for a given K. With BAPS, 10 repetitions for each K = 10, 20 and 30 as maximum bounds of the numbers of clusters were carried out. The results of the mixture analysis served as input for BAPS admixture analysis based on 500 simulations from posterior allele frequencies. A network of clusters where gene flow is indicated by weighted arrows, such that the weights equal relative average amounts of ancestry in the source cluster among the individuals assigned to the target cluster was estimated as described by Tang, Hanage, Fraser, and Corander (2009) using the function Plot Gene Flow in BAPS.
We used analysis of molecular variance (Excoffier, Smouse, & Quattro, 1992) as implemented in GenAlEx version 6.501 (Peakall & Smouse, 2006, 2012 to estimate the partitioning of genetic variance among taxa, among populations and within populations. We determined significance with 9,999 permutations.

| Mitochondrial gene tree and sequence diversity
Maximum-likelihood analyses of cox1 sequences (655 positions) of the Alopia taxa and one Albinaria puella puella and one Herilla ziegleri dacica as outgroups showed that the Alopia specimens from the Bucegi Mountains form three strongly supported clades, representing A. pomatias, A. straminicollis and the A. livida group including A. fussi and A. nixa (Figure 1). Within A. pomatias p-distances reached 0.6%, within the A. straminicollis clade 5.3% and within the A. livida clade 4.2%; between the three clades pdistances reached 10.1% (Table S3).
The relationships between the three clades were not robustly resolved. Within the A. livida group, only the A. nixa specimens formed a well-supported monophylum, whereas individuals identified as A. livida nubila, A. livida bipalatalis, A. livida hypula and A. fussi intermingle. In the A. straminicollis clade, the individuals from the northern slope of the Bucegi Mountains representing A. s. straminicollis formed the sister group of the individuals from the southern part of the mountains representing A. straminicollis monacha. Among A. straminicollis monacha, there was also a dextral specimen from the hybrid zone with A. livida, identified as A. livida because of its dextral coiling. Likewise, there was a sinistral specimen from the hybrid zone between A. s. straminicollis and A. livida in the A. livida group that was identified as A. s. straminicollis because of its sinistral coiling.

| Network based on AFLP data
Using six primer combinations, we scored 2,215 AFLP fragments of 50-500 bp length in 250 Alopia specimens (Table  S4). In contrast to the mitochondrial gene tree, the individuals belonging to one population usually form a cluster in the neighbour-net based on the nuclear AFLP markers (Figure 2). The only species that is separated from the other species by a long branch is A. pomatias. The other taxa form an almost star-like radiation. The arrangement of the other populations in the network reflects their geographical relationships. At the one side of the network, the array starts with the A. straminicollis monacha populations from the southern part of the Bucegi Mountains (see Figure S1). Then, the hybrid population between A. straminicollis monacha and A. livida nubila (population 2) and the populations of A. livida nubila follow. The next branches are populations from higher altitudes classified as A. livida kimakowiczi and A. fussi. Among the A. fussi populations, the populations of A. nixa form a distinct cluster. Only one individual of A. nixa is separated from this cluster. The populations of A. livida bipalatalis, which are geographically placed between A. fussi and A. livida hypula, are also placed between these taxa in the network. Finally, the hybrid population between A. livida hypula and A. s. straminicollis (population 20) forms the transition to A. s. straminicollis.

| Population genetic structure
The ad hoc quantity ΔK, proposed by Evanno et al. (2005) to estimate the number of clusters, shows a maximum for K = 2 ( Figure S2B). However, a plot of the likelihood of K for K = 1-12 showed that L(K) does not reach a plateau in this range, but that it further increases with higher values ( Figure S2A). Thus, we show the STRUCTURE results for K = 2-8 that give additional insights into the genetic struc-  Encircled numbers refer to sampling localities and colours refer to Alopia taxa (see Figure S1 and Table S1). In all results with K = 8, an additional cluster is found that never reaches a proportion above 75% and does not represent a specific taxon or population group. When K is further increased, the additional clusters do not represent distinct population groups.
The STRUCTURE analysis provides evidence for extensive gene flow between A. straminicollis and A. livida. In the contact zones (2 and 20), all individuals have a mixed ancestry and also in the neighbouring populations several individuals show high genetic proportions of the other species. There is also admixture between the other enantiomorph pair, A. nixa and A. fussi when they form separate clusters with K = 8 (Figure 3). Individuals of populations 6 and 7 F I G U R E 2 Neighbour-net based on Jaccard distances between AFLP data of 250 Alopia individuals from the Bucegi Mountains. Encircled numbers refer to sampling localities and colours refer to Alopia taxa (see Figure S1 and Table S1). Coiling direction is indicated by D = dextral or S = sinistral after the taxon names [Colour figure can be viewed at wileyonlinelibrary.com] The BAPS admixture solution showed less admixture than the STRUCTURE analysis with K = 6 ( Figure 3).

An analysis of gene flow between predefined population groups corresponding to A. pomatias albicostata, A. straminicollis monacha, A. livida nubila, A. livida kimakowiczi, A. nixa, A. fussi, A. livida bipalatalis, A. livida hypula and
A. s. straminicollis using a gene flow plot calculated with BAPS ( Figure S3) indicated that gene flow downhill is distinctly higher than uphill in four of five cases in which gene flow exceeds 1% at least in one direction (Table S5). The exception is a higher amount of gene flow from A. livida nubila to A. fussi at higher altitudes.
An analysis of molecular variance (Table S6) attributed only a small part of the genetic variation (13%) to the division into five species proposed by Nordsieck (2008Nordsieck ( , 2007Nordsieck ( , 2016. The variation between populations accounted for 26% and the variation within populations for 61% of the total variation.

| Stages of the Alopia radiation
Alopia in the Bucegi Mountains offers the opportunity to study all stages of a radiation from slight differentiation of populations to complete speciation within a few kilometres. We will discuss the different stages beginning from the case of complete speciation. Alopia pomatias is monophyletic in the mitochondrial gene tree (Figure 1), is distinctly separated from all other taxa in the neighbour-net based on nuclear markers ( Figure 2) and shows little admixture with other taxa (Figure 3). It is the only taxon in the Bucegi Mountains belonging to clade F in the mitochondrial gene tree of Fehér et al. (2013). This clade corresponds to the subgenus Kimakowiczia that differs from the other Alopia in elongated male organs and an elongated diverticulum of the bursa copulatrix (Grossu, 1981;Nordsieck, 2008). Nordsieck (2016) reported a possible hybrid between A. pomatias and A. fussi and raised the question whether the elongation of the male organs of A. nixa and of A. fussi (Grossu, 1981;Szekeres, 1976) might be the result of introgression. In any case, A. pomatias is the only taxon in the Bucegi Mountains that co-occurs with other taxa without fusing with them. At each of its localities, it is associated with a species of the A. livida clade, either A. nixa, A. fussi or A. livida kimakowiczi. This is the strongest evidence that it represents a distinct species. In A. nixa, the speciation process is less advanced than in A. pomatias. Besides A. pomatias, this is the only taxon in the Bucegi Mountains that is monophyletic in the mitochondrial gene tree (Figure 1) and forms a distinct cluster in the neighbour-net (Figure 2; one specimen separated). Alopia nixa differs from A. livida and, to a lesser degree, from A. fussi in elongated male and female organs (Grossu, 1981;Szekeres, 1976) and in the sinistral coiling. Since the coiling direction affects copulation, it is assumed that a change in coiling direction results in assortative mating and frequency-dependent selection against immigrants with the opposite coiling direction (Asami et al., 1998;Johnson, 1982;Ueshima & Asami, 2003). This might have contributed to the differentiation of A. nixa. However, the admixture analysis ( Figure 3) and the gene flow plot ( Figure S3) indicated admixture with A. fussi, the most closely related taxon, which resembles A. nixa morphologically with exception of the coiling direction, and, to a lesser extent, with A. livida. The finding of a co-occurrence of A. nixa and A. livida bipalatalis with a high number of morphologically discernible hybrids by Nordsieck (2016) indicates that the observed admixture between A. nixa, A. fussi (potential hybrids with which are morphologically not discernible) and A. livida can at least partly be ascribed to ongoing gene flow between these taxa. The monophyly of the haplotypes of A. nixa in the mitochondrial gene tree might indicate that there is selection against mitochondrial introgression. In the unconstrained BAPS analysis (Figure 3), A. nixa and A. fussi were not separated. More detailed analyses of the contact zone between A. nixa and A. fussi are necessary to estimate the amount of current gene flow and the constancy of the differences in the genitalia between these two taxa. Based on the available data, we suggest separating A. nixa and A. fussi only as subspecies (see also remarks on the classification below) in accordance with Fehér et al. (2013) who classified A. fussi as subspecies of A. nixa.
The A. straminicollis clade (clade C2 in the mitochondrial gene tree of Fehér et al., 2013), which includes the populations from the northern slope of the Bucegi Mountains classified as A. s. straminicollis and the populations from the southern part classified as A. straminicollis monacha may be considered an example of "failed speciation." The mitochondrial gene tree (Figure 1; Fehér et al., 2013) showed that this clade separated from the branch leading to the A. livida complex in an early stage of the Alopia radiation. Alopia s. straminicollis and A. straminicollis monacha differ from the taxa of the A. livida complex except A. nixa in the sinistral coiling. Thus, it might be expected that these taxa are in an advanced stage of speciation. However, the gene flow estimates between A. livida nubila and A. straminicollis monacha and between A. livida hypula and A. s. straminicollis are higher than those between the taxa within the A. livida complex ( Figure S3). There is also frequent introgression of mitochondrial haplotypes (Figure 1; Koch et al., 2017). The hybrid zones between A. livida and A. straminicollis in the Tatar Gorge and the Velican Valley were known based on morphological evidence (Nordsieck, 2007;Szekeres, 1976Szekeres, , 2008Szekeres, , 2016. However, Nordsieck (2008Nordsieck ( , 2007 supposed that the two taxa hybridize only to a limited extent. In contrast, the admixture analyses of the nuclear markers ( Figure  3; see Koch et al., 2017 for additional analyses) showed that most specimens in the hybrid zones have a combination of similarly large components from the genomes of the parental taxa irrespective of their classification (Figure 3; Koch et al., 2017). The introgression between the two taxa in the hybrid zones resulted in a fusion of their populations. This is also visible in the neighbour-net in which the individuals of the hybrid populations are not separated according to the classification (Figure 2). The introgression is not limited to the hybrid zones. Apparently "pure" A. s. straminicollis and A. straminicollis monacha populations showed some introgression with genomic components of A. livida as well, albeit the exact pattern depends on the number of assumed clusters (Figure 3). Genes for the opposite coiling direction are probably erased from these populations by frequency-dependent selection as long as there are only a few immigrants with the opposite coiling direction. This results in a mosaic-like geographical pattern of populations of A. livida nubila and A. straminicollis monacha in the southern part of the Bucegi Mountains ( Figure S1; Szekeres, 1976: Figure 2). However, if the number of immigrants increases, for example, because they are regularly washed down from an upstream population, frequency-dependent selection breaks down as observed in the hybrid zones between A. livida and A. straminicollis in the Tatar Gorge and the Velican Valley. It had already been observed by Nordsieck (2016) that the percentage of the hybrid phenotypes depends on the frequencies of the enantiomorph taxa. Nordsieck (2008) argued that dextral and sinistral taxa should be classified as distinct species because of their mosaic-like distribution. This conclusion was also based on the assumption that there is only limited hybridization between these taxa. In contrast, our data showed that the A. straminicollis populations fuse with neighbouring populations of A. livida when in contact and that even apparently "pure" populations showed some introgression (Figures 2 and  3). Thus, A. s. straminicollis and A. straminicollis monacha cannot be separated as a distinct species from A. livida. We suggest classifying them as subspecies of A. livida as it has been proposed by Szekeres (1976).
The dextral taxa of the A. livida complex represent the least differentiated stage of the radiation. The A. livida complex forms a shallow clade in the mitochondrial gene tree in which only the sinistral A. nixa is monophyletic (Figure 1). Alopia livida nubila from the southern part of the Bucegi Mountains and A. livida hypula from the northern slope are placed at different ends of the neighbour-net (Figure 2). In the admixture analyses with BAPS and corresponding STRUCTURE analyses (Figure 3), A. livida nubila and A. livida hypula form distinct clusters. Thus, we suggest classifying these population groups as subspecies despite the lack of morphological differentiation. They are not directly in contact but are connected across the main ridge of the mountains by populations classified as A. fussi and A. livida bipalatalis. The populations usually classified as A. livida bipalatalis include genomic components of the A. nixa + A. fussi cluster and of A. livida hypula (Figure 3). Likewise, the populations from the south-eastern range called here A. livida kimakowiczi for descriptive purposes include genomic components of the A. nixa + A. fussi cluster and of A. livida nubila (Figure 3). Because A. livida bipalatalis and A. livida kimakowiczi do not form separate clusters, we do not consider these transitional groups as distinct subspecies. However, the delimitation of population clusters may depend on the sampling. A representative sampling across the whole range of A. livida will be necessary to achieve a final classification. The transitions between the A. nixa + A. fussi cluster and the neighbouring A. livida subspecies show that A. nixa and A. fussi cannot be separated as distinct species and we suggest classifying them also as subspecies of A. livida. Consequently, the Alopia species of the Bucegi massif are classified as two species, A. pomatias (with the subspecies albicostata and pomatias) and A. livida (with the subspecies fussi, hypula, livida, monacha, nixa, nubila and straminicollis). In the following, we will use this classification (see also Table 1).
A comparison of the p-distances between cox1 sequences within and between Alopia clades with corresponding values of other stylommatophoran radiations (Table S3) confirms that Alopia represents an early stage of radiation (see also Fehér et al., 2013). Although species status should be assessed based on the criteria of the applied species concept and not on distances between marker sequences (see also Davison, Blackie, & Scothern, 2009;Sauer & Hausdorf, 2012), the low distances within and between Alopia clades show that our suggestion to classify most of the taxa in the Bucegi Mountains as subspecies of A. livida is not indiscriminate lumping, but is also justified in comparison with the classification of other stylommatophoran groups.

| Dispersal processes and their effects on diversification
If we want to understand non-adaptive radiations, we must also understand the processes that provide opportunities for speciation and those counteracting differentiation and speciation. Beside differentiation, dispersal processes are most important. The distribution patterns of Alopia species and the admixture between Alopia populations in the Bucegi Mountains indicate that dispersal of Alopia specimens occurs more frequently than expected before. Dispersal processes in land snails can be classified into three categories: active dispersal, passive short-distance dispersal and passive longdistance dispersal. In land snails, active dispersal is generally slow and limited to short distances. In habitat specialists like the rock-dwelling Alopia, it is further limited by the restriction that unsuitable habitat has to be crossed to reach the next suitable habitat island. There are no data about active dispersal in Alopia, but we can assume that Alopia behaves similar as the rock-dwelling Albinaria. Albinaria individuals move actively at most a few metres in their whole life (Schilthuizen & Lombaerts, 1994). Thus, active dispersal might contribute to the coherence of demes of neighbouring rock outcrops, but it is unlikely to act as a homogenizing force at spatial scales larger than a few tens of metres (Schilthuizen & Lombaerts, 1994).
Short-distance passive dispersal is more important from an evolutionary perspective. The main agents of passive dispersal of snails in the mountains are probably streams and heavy rains as well as falling rocks that transport snails downhill. Thus, we expect that the direction of short-distance passive dispersal is biased. The higher gene flow estimates from Alopia populations groups at higher altitudes to neighbouring groups at lower altitudes than vice versa ( Figure S3, Table 1) support the importance of short-distance passive dispersal. There is also evidence for the importance of downhill passive dispersal by streams in other land snails (Arter, 1990). Although it has been shown for Arianta arbustorum that gene flow by passive short-distance dispersal is low and allows the differentiation of local populations (Arter, 1990), it might counteract the speciation process between populations that adapt to different altitudes (e.g. A. livida fussi vs. A. livida hypula and A. livida nubila) in an evolutionary perspective.
The distribution patterns within Alopia provide also evidence for the third dispersal category, long-distance passive dispersal. The phylogenetic analysis of Fehér et al. (2013) revealed or confirmed at least three disjunctions over more than 100 km that cannot be explained as remnants of formerly continuous ranges; for example, the colonization of central Transylvania by A. livida from the Bucegi Mountains. Many disjunctions over shorter distances probably also belong in this category. For example, the populations of A. livida straminicollis at the northern slopes of the Bucegi Mountains and of A. livida monacha in the southern part ( Figure S1), which are sister groups according to the mitochondrial gene tree (Figure 1) were probably not directly connected, but originated also by a long-distance dispersal event. Long-distance dispersal requires other dispersal agents than short-distance passive dispersal. The most likely vehicles for long-distance transport of snails are birds (Gittenberger, Groenenberg, Kokshoorn, & Preece, 2006;Rees, 1965;Simonová, Simon, Kapic, Nehasil, & Horsák, 2016;Wada, Kawakami, & Chiba, 2012). In contrast to the discussed short-distance passive dispersal, successful transport of living snails by birds is usually undirected (though major migration routes of birds may influence its likelihood) and much rarer. Whereas short-distance passive dispersal may counteract differentiation of neighbouring populations, long-distance dispersal provides new opportunities for speciation because it results in geographically isolated populations that can differentiate without gene flow. Moreover, only few individuals are involved in long-distance dispersal events resulting in founder events that further promote speciation (Gavrilets & Hastings, 1996;Mayr, 1963).
The opposite effects of the different modes of dispersal on the generation of biodiversity are nicely shown by the taxa of the clade C2 in the mitochondrial gene tree of Fehér et al. (2013)

| Non-adaptive radiations versus polytypic species
Species may be kept separate by reproductive barriers and/ or by disruptive natural selection resulting from differential adaptation to different environments. Disruptive natural selection often maintains the distinctness of species long before reproductive isolation is complete (Wu, 2001). However, in a non-adaptive radiation, disruptive natural selection is lacking. Thus, reproductive barriers may evolve only by sexual selection or by random genetic drift in isolated populations. Genetic drift is expected to be slow so that the waiting time to speciation may last much longer if no selection is involved (Gavrilets, 2003). In the case of the Alopia taxa in the Bucegi Mountains, this is demonstrated by the fusion of populations belonging to the distantly related clades C2 (including A. livida straminicollis and A. livida monacha) and D1 (including A. livida sensu stricto) of the mitochondrial gene tree of Fehér et al. (2013) in the southern part as well as on the northern slopes of the mountains. Despite the long separation of these lineages (Figure 1; Fehér et al., 2013), no sufficient barriers evolved so that extensive gene flow between the northern populations of the two lineages, A. livida straminicollis and A. livida hypula, as well as between the southern lineages, A. livida monacha and A. livida nubila, initiated the fusion of these taxa. The geographical adjacent representatives of the two lineages became more similar to each other on the genetic level than they are to the populations of the same clade inhabiting the opposite region of the Bucegi Mountains (Figures 2 and 5).
Morphological disparity between populations may increase even if they are connected by limited gene flow. However, increasing morphological disparity is not equivalent to speciation. Even if we do not require reproductive isolation as in the biological species concept (Mayr, 1942(Mayr, , 1963, but accept the differential fitness species concept (Hausdorf, 2011), species have to be characterized by features that would have negative fitness effects in other groups. In a non-adaptive radiation, these features cannot be adaptations to different ecological niches, but can only be features affecting reproduction itself or the intrinsic fitness of the offspring. Groups of populations which are neither adapted to different environmental conditions nor have characteristics that restrict gene flow to a level that prevents the fusion of the populations upon contact, cannot be considered distinct species irrespective of the morphological or genetic disparity they display and irrespective of the time they were separated. Rather, they have to be classified as polytypic species. The definition of radiation usually includes "the evolution of a relatively large, monophyletic group of species" (Gittenberger, 1991), the divergence of "a single ancestor … into a host of species" (Schluter, 2000) or "a pattern of species diversification" (Rundell & Price, 2009). The differentiation of subspecies may be the first step in a radiation process. However, as such it is not yet radiation. Particularly in taxa like land snails in which there is little gene flow between populations because of their low dispersal capability, disparity between populations can increase considerably despite a lack of intrinsic characteristics that restrict gene flow. The disparity between populations may exceed that between distinct species, which evolved more quickly as a result of disruptive selection.
The fusions of A. livida straminicollis with A. livida nubila and of A. livida monacha with A. livida hypula show that the time since isolation or the position in a gene tree is irrelevant for the decision whether a group of populations represents a subspecies of a polytypic species or a distinct species. Decisive is only whether the taxon is characterized by features that would have negative fitness effects in other groups. The taxa that are located in the mitochondrial gene tree between the clades C2 (including A. livida straminicollis and A. livida monacha) and D1 (including A. livida sensu stricto) (Fehér et al., 2013) are not necessarily all subspecies of a single polytypic species. If it can be shown that they have characteristics that restrict gene flow with other population groups to a level that prevents the fusion of these groups upon contact, they should be classified as distinct species. Our results challenge the uncritical classification of taxa that have low dispersal abilities but show high disparity between populations as non-adaptive radiations. We need to prove that such taxa are not just polytypic species and we need to investigate the conditions under which polytypic species evolve into radiations despite niche conservation and how frequently they become extinct or fuse with each other before they reach species level (as in the cases of A. livida straminicollis and A. livida nubila or A. livida monacha and A. livida hypula).
The radiation of the land snail genus Xerocrassa in Crete illustrated how non-adaptive radiation can originate despite of the lack of disruptive natural selection and the long waiting time for speciation based on genetic drift alone by sexual selection (Sauer & Hausdorf, 2009). Sexual selection may increase the rate of reproductive divergence between populations and thereby drive speciation (Dominey, 1984;Gavrilets, 2000Gavrilets, , 2014Panhuis, Butlin, Zuk, & Tregenza, 2001;Ritchie, 2007). In adaptive radiations, it may act in concert with divergent natural selection. In contrast, it is the only mechanism that can speed up the diversification process beyond the slow pace possible with genetic drift alone that will rarely result in radiation, the rapid diversification of a lineage into multiple species. The fact that the Alopia taxa that reached the most advanced stage of speciation, that is A. pomatias, shows the most distinct differences in the genitalia indicates that sexual selection has also facilitated the radiation of Alopia. In A. pomatias and, to a lesser degree, in A. livida nixa and A. livida fussi, the epiphallus that produces the spermatophore is elongated compared to A. livida straminicollis and A. l. livida (Grossu, 1981;Szekeres, 1976). In stylommatophoran snails, the spermatophore is transferred into the bursa copulatrix complex during the copulation, where the spermatophore with most of the sperm is digested (Lind, 1973;Rogers & Chase, 2001). An elongation of the spermatophore-producing organs, the epiphallus and the flagellum (if present) results in longer spermatophores that increase the number of sperm cells that can leave the spermatophore before it is introduced into the bursa copulatrix complex and digested. Thus, there is a sexual conflict between the male partner that wants to fertilize as many eggs as possible and the female partner that wants to control the fertilization of the eggs. This conflict can result in a co-evolutionary arms race between spermatophore-producing and spermatophore-receiving organs (Koene & Schulenburg, 2005;Sauer & Hausdorf, 2009). The elongation of the spermatophore-producing epiphallus and of the diverticulum of the bursa copulatrix, the spermatophore-receiving-organ, in the Alopia taxa that represent the most advanced stages of speciation in the radiation, A. pomatias and A. livida nixa, indicates that also the Alopia radiation was driven by sexual selection. This further supports the general importance of sexual selection in non-adaptive radiations.