Influence of fruit dispersal on genotypic diversity and migration rates of a clonal cactus from the Chihuahuan Desert

Abstract The diverse offspring of clonal species differ in their dispersability, influencing genotypic diversity and clonal structure. Here, we determined dispersal patterns and their impact on genetic structure in Opuntia microdasys, a self‐incompatible cactus with three dispersal units (one sexual and two clonal). We analyzed dispersal, using experiments at three populations, and assessed multilocus genotypes (ISSR markers) of all individuals in 10 clumps per population with known reproductive origin (sexual or clonal). Genotype of all samples, population structure, and migration between clumps and populations were assessed with GenAlEx and GenoDive, assuming higher genotypic diversity and migration when sexual reproduction is more frequent. We determined the most likely number of genetic clusters with STRUCTURE and geneland. Dispersal differed among populations; primary dispersal occurred at short distances and was farthest on steep slopes, and dispersal distance increased after secondary dispersal. Clumps had 116 different multilocus genotypes in three spatially explicit genetic clusters. We detected genetic structure at small scale, genotypic diversity among clumps varied between populations; diversity decreased while clonal dominance increased, and the most variation occurred among clumps. Genetic structure was moderate, suggesting gene flow by seed dispersal allows slight differentiation among population at large scales. Genetic diversity within clumps was the lowest because dispersal of clonal propagules was limited and caused genotypic dominance at local scale. However, the combined dispersal pattern of sexual and clonal dispersal units is fine‐tuned by environmental factors, generating a range of genetic diversity among clusters and populations. This pattern suggests that genetic structure of clonal plants is more dynamic than thought, and dispersal of different types of offspring affects genetic structure at many scales.

clonal propagules often lack specialized dispersal structures, dispersal is assumed to be limited (Bullock et al., 2006;Eckert, 2002;Winkler & Fischer, 2002). In species with mixed reproduction (i.e., combined sexual and clonal recruitment), the spatial genetic structure has two opposite patterns (Alberto et al., 2005), either dispersal and subsequent establishment promote the spatial arrangement of intermingled ramets of different genets (i.e., multiclonal patches) or limited dispersal of clonal propagules (Bobich & Nobel, 2001;Bravo-Hollis, 1978;Fuentes Pérez, 2008;Negron-Ortiz & Strittmatter, 2004;Nobel, 2002;Piña et al., 2007) leads to groups of clumped ramets of the same genet (i.e., genotypic dominance in monoclonal stands [superclones]) (Alberto et al., 2005;Barrett, 2015;Charpentier, 2002;Gélin et al., 2017). On one hand, dispersal by direct observation is plausible for species with large and easily traceable dispersal units; for these cases, mark and track experiments are useful to determine the source of clonal propagules. Although direct methods provide exact information on dispersal distances, the difficult task of gathering data for dispersal over long distances poses a serious limitation (Bullock et al., 2006;Nathan & Muller-Landau, 2000;Nathan, Perry, Cronin, Strand, & Cain, 2003). In addition, in species with high clonal recruitment, it is hard to determine the source of a ramet, as some species tend to be dominated by a superclone (Bravo-Hollis, 1978). On the other hand, indirect methods that use molecular markers are well developed to evaluate effective dispersals (i.e., dispersal plus establishment events; Cain, Milligan, & Strand, 2000, Levin, Muller-Landau, Nathan, & Chave, 2003 and determine the number and distance of migrants per generation and the degree of genetic structure and differentiation between populations (Cain, Milligan, & Strand, 2000;Carrillo-Angeles et al., 2011;Levin, et al., 2003;Manel, Gaggiotti, & Waples, 2005;Pritchard, Stephens, & Donnelly, 2000a). Indirect methods, however, are focused on effective dispersal (Cain et al., 2000) and exclude all the reproductive structures that dispersed but have not established or survived.
Our study species, Opuntia microdasys (Cactaceae), is a clonal cactus that produces three kinds of offspring, one of sexual origin (seedlings from seeds formed from ripe fruits: sexual diaspores) and two of clonal origin (detached cladodes that take root and unripe fruits that can form new plantlet recruits; Palleiro et al., 2006). Frequent short-distance dispersal of clonal diaspores will result in spatial aggregation of clone mates (Bobich & Nobel, 2001;Bravo-Hollis, 1978;Fuentes Pérez, 2008;Negron-Ortiz & Strittmatter, 2004;Nobel, 2002;Piña et al., 2007). Morphological and demographic differences among clonal and sexual diaspores of Opuntia microdasys provide an interesting model to assess the dispersal of sexual and clonal diaspores. Palleiro et al. (2006) found that the plantlets mainly establish under the canopy of adults individuals no more than ca. 90 cm from the parent, forming clusters of new offspring under the canopy of adults plants (i.e., clumps of plants).
But not only clusters of clonal propagules become established; Dean and Milton (2000) found clusters of intermingled genets of Opuntia ficus-indica around telegraph poles and wire fences from seeds dispersed by crows. The demographic contributions of each type of propagule (Palleiro et al., 2006) and the spatial configuration of genotypes  produced a gradient of clonality and sexuality between populations. Thus, we expected that dispersal promotes the intermingling of ramets of different genets (i.e., multiclonal patches) and higher genetic diversity in the more sexual population and monoclonal clumps with low genetic variation in less sexual populations where clonal diaspores remain in close proximity. Evaluating fruit dispersal should help elucidate whether seeds (ripe fruits) or seedless (unripe) fruits move longer distances from parent plants. In addition, in species with high clonal recruitment, it is hard to determine the source of a ramet, as some species tend to be dominated by a superclone (i.e., over-representation of ramets with the same multilocus genotype) (Bravo-Hollis, 1978). Because the interaction between the environment, dispersal availability, and type of dispersal unit imposes a challenge when studying dispersal of a clonal species, we combined direct and indirect methods to assess dispersal patterns of sexual and clonal dispersal units (Bullock et al., 2006;Nathan et al., 2003).
Here, we aimed (a) to determine the spatial genetic structure that results from dispersal and establishment events of either sexual diaspores or clonal propagules and (b) to determine the genotypic diversity and migration rate within and between both clumps of plants and populations of O. microdasys (Cactaceae) in the southern Chihuahuan Desert.

| Study species
Opuntia microdasys (Lehm.) Pfeiff. (Cactaceae; Figure 1a), bunny ears or blinding prickly pear, is a self-incompatible, clonal cactus that forms shrubs up to 1 m tall, with oval, bright green cladodes (racket-like stems) that lack spines (Bravo-Hollis, 1978). Areoles have numerous reddish brown or yellow glochids. The segments of the perianth in the flowers are yellow, with flowering between April and May. The fruits are globose, fleshy, 2-2.5 cm in diameter, and turn from green into red when mature, usually ripen between June and August. Unripe and mature fruits either disperse by gravity or are removed by birds and mammals (E. García-Morales, personal field observations). When the fruits reach the ground, several factors could influence their dispersal, but the immediate factors are the slope and microtopography of the site. Other agents such as temporary streams can move fruits on the ground farther during a heavy rain (M. Mandujano, unpublished data).
Opuntia microdasys usually grows on sandy to loamy calcareous soils in hills and uplands in the Chihuahuan Desert of Mexico.

| Fruit dispersal
Two field experiments were set up to explore fruit dispersal of O.
microdasys during fruiting. In the first experiment, we measured primary dispersal, the distance traveled by fruits immediately after they had detached from the parent plant. In the second experiment, we measured the combined effect of primary and secondary dispersal during the entire reproductive season by marking and tracking the fruits.

| Primary fruit dispersal
We experimentally simulated fruit dropping from parent plants at each population (BH, HPH, and IDH) to assess the process of pri- we measured the distance traveled by the fruit (cm) and recorded the quadrant in which the fruit was located (fate). We analyzed the fruit fate with circular statistics (Fisher, 1996). Correlation between the number of fruits that fell in each quadrant (circular data) and distance reached by fruits (linear data) was analyzed using Oriana v4.02 (Kovach, 2011). Finally, the dispersal distance of fruits was analyzed with a two-way nested ANOVA (Kutner, Nachtsheim, Neter, & Li, 1996) in JMP 8.0.2 (population and quadrants as fixed factors and focal individual nested within population as random effect) and a Tukey test (Kutner et al., 1996).

| Marking and tracking fruits
The fate of fruits was followed over the entire reproductive season (from June to September of 2010) to assess the distances reached by fruits over longer periods. In this experiment, the distance from the parent plant to the point where the fruits were found could have resulted from primary and/or secondary dispersal. Here, we assumed that any fruits found beyond the mean distance from the focal plant populations, and month as factors (Kutner et al., 1996). Finally, the number of fruits for each fate was analyzed using a generalized linear model with a Poisson distribution of residuals and the log link function in JMP 8.0.2; fruit count was the response variable with fates of fruits, populations, and sampling period as fixed factors (Kutner et al., 1996).

| Sample collection
We  Table 1). Approximately 10 g of fresh tissue from newly produced cladodes was collected from all ramets (physiologically independent individuals). This tissue proved to be most suitable for extraction and amplification of DNA. Each sample was placed in a sterile 5-cm polyethylene bag with 5 g of silica gel. The silica gel was changed periodically until the tissue was completely dry (this process is required because the Opuntia tissue contains mucilage that interferes with DNA extraction). DNA was extracted with a Fast-DNA Kit (116540600 MP Biomedicals), and we test the quantity and purity with electrophoresis in a 2% agarose gel to obtain between 10 and 20 ng.

The use of dominant markers (such as ISSRs or AFLPs) is a common technique used in ecological and systematic studies of plants and
other organisms because of their low cost and high reproducibility, variable loci, and distribution throughout the genome (Bornet & Branchard, 2001;Nybom, 2004;Zietkiewicz et al., 1994). Compared with other dominant markers such as RAPDs, ISSRs are advantageous for two reasons: High annealing temperatures in ISSR protocols make PCR conditions more stringent for the amplification of fragments (Nybom, 2004), and the longer primers seem to provide the same reproducibility as microsatellites (Bornet & Branchard, 2001;Nybom, 2004). Preliminary tests were done to standardize the protocols for several primers, and three primers (817 [  threshold to assign multilocus genotypes of three bands, Figure 3, see below). The error rate estimated for this study was 3.15%. We also checked that no band exceeded a frequency of 1 − (3/N) according to the proposal of Lynch & Milligan (1994) for dominant markers.

| Genetic data analysis
When molecular markers are used to define multilocus genotypes To avoid this bias, we used the program GenoDive (Meirmans & Van Tienderen, 2004), which plots the differences between individuals against the frequency of individuals. A multimodal histogram frequently results, and a threshold must be chosen to determine the number of clonal lineages in the sample (Douhovnikoff & Dodd, 2003). For choosing a threshold for genetic differences among pairs of individuals and excluding scoring errors or small differences due to somatic mutations, Meirmans and Van Tienderen (2004) recommended using the valley between the first and the second peak as the threshold. In the present study, we thus set the threshold for determining differences at three bands ( Figure 3).
We determined the probability that the detected genotypes are unique genets using P dgen of Sydes and Peakall (1998), which is the probability of drawing a second copy of a particular genotype, given that one copy of this genotype has already been drawn from the population and assuming a population with sexual reproduction and random mating. P dgen can be extended to the probability of drawing the same genotype n times as (P dgen ) n-1 . Further, P dgen = ∏ p i , where For the total sample and for each clump of plants, we calculated the percentage of polymorphic bands (%P) and the number of private bands (PB) with the program GenALEX 6.4 (Peakall & Smouse, 2006. We also calculated the genotypic diversity index corrected for sample size: where G is the number of genotypes identified in the sample and n is the sample size. R can have values between 0 and 1; R is 0 when all individuals are copies of the same genotype and 1 when all individuals have different genotypes (Dorken & Eckert, 2001).
An AMOVA for diploid binary data was used to determine the distribution of genetic variation among the levels of organization included in the study, clumps of plants and populations. This method calculates a matrix of Euclidean distances between pairs of individuals (Excoffier, Smouse, & Quattro, 1992). This test was performed with GenALEX 6.4 (Peakall & Smouse, 2006 after running 1,000 iterations.
We calculated Φ PT , which refers to a relation of the genetic variance among the populations relative to the total variance, but based on information of differences, matrix and differentiation among populations were calculated via AMOVA from haplotypes or dominant markers (Excoffier et al., 1992;Peakall & Smouse, 2006

| Primary fruit dispersal
The mean directional angle (μ), mean resultant length of dispersal (r), and the chi-squared test for uniformity at which fruits fell in the primary dispersal experiment differed between populations ( Table 2). The slope of each habitat affects the spatial pattern of fallen fruits. More fruits that dropped in the west quadrants fell with the highest frequencies to the SW, NW, and W at HPH (Appendix 1a), which resulted in a μ close to a west orientation ( mean direction approximately to the E (Table 2 and Appendix 1c).
The IDH population had the lowest frequency of fruits with a skewed orientation around a focal plant (r = 0.03, κ = 0.061, and SD = 151.49°).
The correlation between direction and fruits dispersal distance was analyzed with circular-linear correlations and was significant for all populations (HPH, r = 0.393, p < 0.0001; BH, r = 0.207, p < 0.0001; IDH, r = 0.107, p < 0.0001). The mean dispersal distance of fruits differed between populations (ANOVA, F 2,27 = 9.53, p < 0.001) and quadrants (F 7,9556 = 69.31, p < 0.001), and the interaction between factors was significant (F 14,9556 = 67.31, p < 0.001). The interaction reflects the influence of specific quadrants in each habitat; for example, quadrants with west orientations determined the fate of fruits in HPH (Appendix 2). The longest mean dispersal distance reached by fruits occurred in HPH (58.4 cm), followed by 32.18 cm in IDH, and the minimum distance was found in BH (29.9 cm; Appendix 2).

| Marking and tracking fruits
The  (χ 2 = 4.1, df = 3, p = 0.2491) and stage of fruit maturity (χ 2 = 2.0, df = 2, p = 0.3673) were not statistically significant, but month had a significant effect (χ 2 = 194.4, df = 3, p = <0.0001) as fruits mature (Appendix 3a). The interaction of population with the maturity stage of fallen fruits was statistically significant (χ 2 = 22.01, df = 6, p < 0.0001), because most fruits fell in the first or second count at BH and IDH and abortion rate was constant over time at HPH. Also, the interaction of the month and the stage of maturity of fruits differed (χ 2 = 65.31, df = 9, p < 0.0001) because the highest proportion of fallen ripe fruits peaked in the second or the third count, unripe fruits peaked in the first count and that proportion decreased toward midseason (Appendix 3a). One other difference is that in IDH, numerous fruits were not found in the first and the last count (Appendix 3a).
We found significant differences in the proportion of fruits with different fates (full model goodness of fit Pearson value:

| Genetic structure of clumps of plants and populations
Of the 577 individual ramets that were genotyped with the three selected ISSR primers (Table 1), we identified 115 different clonal lineages (genotypes) with GenoDive using a three-band threshold of differences between genotypes (Figure 3).
In most clumps of plants, we found clonal individuals. The population with the highest genotypic diversity was HPH (R = 0.435), followed by IDH (R = 0.222) and BH (R = 0.087; Table 1). Plant clump IDH8 had the highest genotypic diversity (R = 1) and the fewest members (only three ramets). In HPH, the highest values of R corresponded to two clumps of plants (HPH3 and HPH6; R = 0.917 and 0.719, respectively; Table 1). On the other hand, in BH and IDH, most clumps of plants had lower genotypic diversity (0 in several cases), indicating a predominantly clonal composition (Table 1). The values of P dgen were very low for all populations (BH = 0, HPH = 1.6e −38 , IDH = 5.13e −89 ), which suggests that the assignment of clones is robust.
In agreement with HPH having the highest genotypic diversity, HPH also had the highest percentage of polymorphic bands and private bands (%P = 82.44%, PB = 23), followed by IDH (%P = 65.95%, PB = 22) and BH (%P = 57.71%, PB = 7,  (Table 3) and only 21% of the variation between habitats. This result means that the differentiation of populations was moderate to high (Φ PT = 0.21). The AMOVA that included the clumps and populations also highlighted the fact that almost all the variation was found between clumps (81%, Φ PT = 0.92, Table 3), with very little variation between populations (11% of variation) or within clumps (8%).
We tested two Bayesian methods of assignment of individuals into genetic clusters: STRUCTURE (Pritchard, Stephens, & Donelly, 2000b) and GENELAND (Guillot et al., 2005). The same number of clusters (K = 3) was chosen in both analysis, but assignment of individuals differed between the two methods. The three clusters predicted by STRUCTURE were represented by individuals from all three populations in different proportions. In BH, the proportion of individuals in cluster one (red cluster in Figure 5a) was 0.99; in the HPH, three genetic clusters were present in similar proportions (red cluster = 0.4, green cluster = 0.37, and blue cluster = 0.23, Figure 5a); and in the IDH genetic clusters, one and three were predominant (red cluster = 0.55 and blue cluster = 0.44, Figure 5a). In the GENELAND analysis, the genetic clusters corresponded closely to the spatial distribution of individuals.
All individuals from IDH were assigned to one cluster (white cluster in Figure 5b). All individuals but one (BH1) from BH were assigned to another cluster (tan cluster in Figure 5b), and all individuals from HPH, with the BH1, were assigned to a third cluster (green cluster in Figure 5b).  to determine the consequences of dispersal on genotypic diversity for clumps of plants and between populations.

| D ISCUSS I ON
On one hand, because plantlets are easily traceable dispersal units, direct observation of dispersal was plausible for this species and gave us an accurate estimate of short dispersal distances, but the estimates were not accurate for long distances (Bullock et al., 2006;Nathan & Muller-Landau, 2000;Nathan et al., 2003).
However, we found other abiotic factors that affected dispersal.
For example, microtopography modified the trajectory and distance reached by fruits, at the population with the steepest slope, fruits reached longer distances and produce clumps of plants with intermingled genotypes (HPH). In contrast, in the population with the shallowest slope (IDH), the fruits accumulated near the source and produced monoclonal clumps. Movement of clonal propagules is influenced by gravity and slope (Chambers & MacMahon, 1994). Few fruits moved longer distances (i.e., 100 m), following the same leptokurtic curve proposed for seed dispersal (Nathan & Muller-Landau, 2000;Willson, 1993). Furthermore, during primary dispersal, fruits move on average only 1 m, even under the influence of a slope, a common phenomenon in species dispersed by gravity (Chambers & MacMahon, 1994;Nanami et al., 1999;Pairon et al., 2006). The influence of orientation, slope angle, and gravity on the formation of clone or seed clumps has been previously quantified (Pairon et al., 2006). For example, Podocarpus nagi seeds forms clumps under the canopies of large female trees (Nanami et al., 1999), and 95% of the seeds from Prunus serotina fall within 0-5 m of the source (Pairon et al., 2006).
The importance of abiotic factors such as river water flow and the hydrologic regime determines the site of deposition of postrelease propagules of Betula fontinalis (Merritt & Wohl, 2002). Shortdistance dispersal deposits the propagules where the offspring will establish; Palleiro (2001) commonly found offspring within a 90-cm radius under the crown of O. microdasys individuals. We expected that the morphological traits of the fruits should confer higher mobility, but this type of propagule shares some traits with other clonal propagules: They are larger than seeds, have shorter dormancy periods and lack a specialized dispersal mechanism, which usually leads to a clumped distribution (Eckert, 2002). Nonetheless, secondary agents increased the mean dispersal distance reached by fruits. We  Interdune   IDH1  IDH2  IDH3  IDH4  IDH5  IDH6  IDH7  IDH8  IDH9  Mandujano, unpublished data). Propagules of species that undergo secondary dispersal are commonly redistributed by a second agent (Bohning-Gaese, Gaese, & Rabemanantsoa, 1999;Griffith & Forseth, 2002), which generally increases the dispersal distance (Nathan & Muller-Landau, 2000). Helsen, Verdyck, Tye, and Van Dongen (2009) suggested that finches carry Opuntia echios seeds long distances between different islands in the Galapagos Islands. Another example of bird dispersal of Opuntia is found in the Karoo, South Africa (Dean & Milton, 2000); crows (Corvus capensis) move seeds of Opuntia ficusindica and may be the most important vector for the range expansion of this Opuntia species.
On the other hand, indirect methods that use molecular markers are well developed and are used to evaluate effective dispersals (i.e., dispersal plus establishment events; Cain et al., 2000, Levin, et al., 2003, determine the number and distance of migrants per generation, and the degree of genetic structure and differentiation of populations (Cain et al., 2000;Carrillo-Angeles et al., 2011;Levin et al., 2003;Manel et al., 2005;Pritchard, Stephens, & Donelly, 2000a).
Indirect methods, however, are focused on effective dispersal (Cain et al., 2000) and exclude all the reproductive structures that dispersed but have not established or survived.
Unlike species with linked clonal growth, in which the ramets are spatially clumped (Charpentier, 2002), O. microdasys produces unlinked ramets from unripe fruits (clonal propagules) with traits that we had expected to provide greater mobility and, consequently, longer dispersal distances. This kind of pseudo-viviparity is a common phenomenon found in other species of Cactaceae such as Opuntia spp. and Cylindropuntia spp. (Bravo-Hollis, 1978;Fuentes Pérez, 2008;Negron-Ortiz & Strittmatter, 2004;Nobel, 2002;Palleiro et al., 2006;Piña et al., 2007;Vázquez-Delfín et al., 2005). Therefore, we expected new plantlets to establish away from the parent plant, resulting in intermingled genotypes. Nevertheless, the response was not straightforward, genotypic diversity in some clumps of plants of BH was very low or even completely clonal, while other clumps were genetically diverse (HPH, Table 1). This pattern reflects the high clonal recruitment that occurs in the clumps of this population and corresponds with the amount of clonality reported by Palleiro et al. (2006). The low genotypic diversity found in IDH could reflect the importance of each phase of dispersal in the genetic configuration of the clumps of plants; actually, the short distance reached by fruits during primary dispersal would favor the formation of clumps of ramets from the same genet, and the fruits that dispersed longer distances or to unsuitable habitats (e.g., Neotoma nests) apparently failed to establish new offspring. The population with the highest genotypic diversity was HPH, for which slope was the main factor influencing primary dispersal, and HPH was the population with the highest percentage of sexual recruitment (Palleiro et al., 2006).
Most genetic variation occurred within a population or among clumps (AMOVA , Table 3); the highest genetic differences were found between clumps, and the value of Φ PT indicated high to moderate genetic differentiation between the studied populations. This pattern is dissimilar to other clonal species with low values of differentiation among populations as in Potamogeton pectinatus in which Abbasi, Afsharzadeh, and Saeidi (2017) found values of Φ PT = 0.11 and the highest genetic variation located within populations (89%).
In a study with Bromus ircutensis, a clonal grass, found F ST values ranged from 0.118 to 0.15% and 87% of the genetic variation within the populations. STRUCTURE analysis assigned some individuals from all three populations to one genetic cluster ( Figure 5a); consequently, we could not recognize a characteristic genotypic pattern for each population, and we found that most of the genotypic variability occurred within populations and between clumps, support- The assignment of genotypes to genetic clusters in STRUCTURE suggests that migration could play a more significant role than we detected with field experiments; in fact, assignation analysis reveals the actual scope of dispersal abilities of propagules and the effective dispersal (Figures 2 and 5). STRUCTURE identified a predominant cluster for all three populations (Figure 5a, cluster in red), but a blue cluster that was present only in IDH and HPH. In contrast, by adding the spatial location of clumps, GENELAND separated the populations, and just one clump of BH was assigned with all clumps of HPH.
The difference between these two analyses could be due to the effect of adding spatial data, as all other factors of organization of this study (i.e., offspring under focal plants, clumps, clumps in quadrants, plots of populations, and populations, Figure 2) were the same.
High migration rates or gene flow either by unripe fruits or seeds could be the reason for the spatial genetic structure; that is, most genetic variation occurred within population and among clumps. In addition, gene flow is also limited by pollen dispersal in Opuntia microdasys, based on the behavior of the primary pollinator (Piña et al., 2007), in addition to either clonal or sexual offspring dispersal. Furthermore, other examples of similar migration rates have been shown for very well-structured populations (Gélin et al., 2017;Yu, Han, Tian, & Liu, 2011) or for small, isolated populations (Kim & Chung, 1995). An extreme example was found in Opuntia echios on the Galapagos Islands, in which no clonal individuals were found in a sample of 444 individuals collected in 22 localities (Helsen, Verdyck, & Van Dongen, 2011). The scale at which levels of variation were observed and the offspring recruitment reported (Palleiro et al., 2006) are evidence indicating a pattern of repeated seedling recruitment (Eriksson, 1992) at the population level, but with clonal recruitment playing an important role at a more local scale (clumps and genet survival).
Although O. microdasys was able to recruit using the three possible pathways in all the studied populations, the percentage of recruitment of each type of offspring differed in each population (Palleiro et al., 2006). This pattern is a clear indicator not only of the capacity of the species to produce clonal propagules and sexual diaspores in a variety of habitats, but also of the ecological conditions that limit the sites where each type of dispersal unit was more successful at establishment and the needs of dispersal units to arrive at these safe sites. The clonal propagules and sexual diaspores must be dispersed to reach these sites (Hroudova & Krahulcova, 1996;Nathan & Muller-Landau, 2000), and the nature of that dispersal determines the pattern of recruitment and distribution of the genetic and genotypic variation in the neighborhood and population.
The reproductive strategy of O. microdasys, maintaining clonal genotypes that can exploit favorable sites in limiting habitats for long periods, often results in monoclonal patches (Gélin et al., 2017;van Groenendael, Klimes, Klimesova, Hendriks, & Van Groenendael, 1996), as occurred in the BH and IDH habitats. However, this strategy incurs a cost by increasing the levels of geitonogamy and limiting the pollen flow between different genotypes (Charpentier, 2002;Zhang & Zhang, 2007). Such pollen limitation has been studied in Maianthemum bifolium, a clonal self-incompatible species, for which fruit set is affected in populations with a low level of genotypic diversity (Honnay, Jacquemyn, Roldán-Ruiz, & Hermy, 2006).
In a self-incompatible species such as O. microdasys (Piña et al., 2007), in which the offspring frequently establish under the parent plant (Palleiro et al., 2006), the aggregation of ramets necessarily leads to some level of geitonogamy (Charpentier, 2002;Zhang & Zhang, 2007), mainly when there is a spatial autocorrelation of genets over short distances (<20 m; Nathan et al., 2003). It is expected that geitonogamy acts as a positive feedback mechanism that favors sexual failure, abortion, and clonality, which thus increase geitonogamy.
The combination of ecological field experiments with molecular genetics experiments allowed us to assess dispersal in a complex system with species with different structures of dispersal and with many levels of organization. Our results suggest that the genetic structure of clumps of plants in part is due to the limited mobility of clonal propagules, bounded by primary dispersal and environmental restrictions for their establishment. This dispersal process leads to an unequal distribution of dispersal unit types at each population (with limited dispersal of clonal propagules and more long-distance dispersal of sexual diaspores), generating monoclonal and intermingled clumps, but with a level of migration over longer distances that allowed some differentiation among populations. The National Council of Science and Technology (CONACyT) provided a scholarship to EGM. We thank P. Vinuesa, B. Milligan, and two anonymous reviewers for comments on the manuscript and Dr. B. E. Hazen for English editing.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
All authors contributed with planning, designing, analyzing, and writing of the manuscript. The study is part of the PhD project of GME and is included within a long-term project of population biology of