Proc. Natl. Acad. Sci. USA Vol. 86, pp. 9044-9047, November 1989 Population Biology Gene flow by pollen into small populations: Data from experimental and natural stands of wild radish (Raphanus sativus/spatial isolation/interpopulation mating) NORMAN C. ELLSTRANDt, B. DEVLINt, AND DIANE L. MARSHALL§ Department of Botany and Plant Sciences and Program in Genetics, University of California, Riverside, CA 92521-0124 Communicated by R. W. Allard, August 14, 1989 (received for review August 18, 1988)

ABSTRACT Gene flow can have an especially strong im- much from population to population, whether it varies within pact on the evolution of small populations. However, empirical a population over a season, and whether it varies with spatial studies on the actual rates and patterns of gene flow into small isolation from the nearest conspecific. More data are needed populations are few. Thus, we sought to measure gene flow into that focus more closely on a number ofpopulations ofa single small populations of wild radish, Raphanus sativus. We found species. significant differences in gene flow receipt among experimental Therefore, we selected wild radish, Raphanus sativus L. populations and within those populations over a season. A (Brassicaceae), to measure patterns of gene flow by pollen maximum-likelihood estimate revealed that almost all of the into small populations. This species is especially suitable for gene flow into these synthetic populations had its origin in both experimental and descriptive gene flow studies because relatively distant (>650 m), large natural populations rather it is a common outcrossing weed in southern California (8), than the proximal (255400 m), small synthetic populations. is easy to grow and transplant (9), is polymorphic for several We also estimated rates of interpopulation mating from simple isozyme loci that are expressed in both adult and seedling paternity analysis of progeny produced by seven small (ca. 50 tissues (10), and has been the subject of prior gene flow study plants) natural populations. Again, we found significant het- (11, 12). First, we used three small, genetically structured, erogeneity in gene flow receipt. Although these populations synthetic populations to measure interpopulation mating varied 10-fold in their range of isolation distances (100-1000 rates and to measure the gene flow coming into these popu- m), gene flow rates did not vary with distance. The magnitude lations from more distant, large natural populations. We of gene flow rates estimated in all but one population was great sampled from these populations periodically to ask whether enough for gene flow to play an important role in the evolution gene flow varies over time within a season. Second, we of these small populations. measured gene flow received by seven small natural popu- lations and asked whether the rates measured varied with the Gene flow has considerable potential as an evolutionary populations' spatial isolation. In both studies, we could not force. Theory indicates that even a moderate rate of gene find any simple relationship of gene flow by pollen to dis- flow is sufficient to counteract the diversifying effects of tance, implying that, for the spatial scale studied, factors local, directional selection, or genetic drift (1). Gene flow can other than distance are more important in determining this have an especially strong impact on the evolution of small critical evolutionary force. populations. Immigration of a few individuals will counteract selection more readily in a small population than a large one MATERIALS AND METHODS (2). Furthermore, although drift becomes more important in smaller populations, a single immigrant per generation will Studies of Synthetic Populations. In February 1987, we suffice to prevent drift (3). created three experimental, genetically structured popula- Not only is gene flow important to the microevolutionary tions of wild radish at the Moreno Valley Field Station of the dynamics of small populations, but such populations are also University of California at Riverside. Natural populations of thought to be especially vulnerable to gene flow by pollen wild radishes occurred at every edge of the Field Station, but because of decreased opportunities for short-range mating were >650 m from each of the experimental populations. To and increased opportunities for receiving the fixed amount of obtain our experimental plants, we electrophoretically pollen arriving from outside the population (4). For example, screened seedlings derived from seed collected at the Field experiments with crop plants have demonstrated that the rate Station in 1986. The seedlings were screened at the following ofgene flow by pollen from source plants bearing a dominant loci: triose phosphate isomerase (TPI), acid phosphatase marker increases as the size of the recipient population (ACP), and two phosphoglucomutase loci (PGMI and decreases (reviewed in ref. 4). Much less is known of rates of PGM2). We obtained allele frequencies of the pollen pool of gene flow into small populations ofnatural species. A handful nearby natural populations from a sample of 320 plants. of studies have used paternity analysis and related methods Details of the electrophoretic procedures have been reported to estimate the rate of gene flow by pollen into small (10). populations (reviewed in ref. 5). These studies reported All experimental individuals were chosen to be monomor- considerable variation in the amount of gene flow detected, phic at the TPI and ACP loci for the same allele. The from as low as 0.1% in Picea glauca (6) to as high as 88% in frequencies of these alleles in the surrounding natural pop- Pinus elliottii (7). However, since almost all of these studies ulations were 0.951 for the TPI allele and 0.824 for the ACP reported on only one or two populations, they provide little allele. These loci are not linked (10). insight into typical patterns of gene flow into small popula- tions for these species, such as whether gene flow varies tTo whom reprint requests should be addressed. tPresent address: Division of Biostatistics, Laboratory of Epidemi- ology and Public Health, Yale University, P.O. Box 3333, New The publication costs of this article were defrayed in part by page charge Haven, CT 06510. payment. This article must therefore be hereby marked "advertisement" §Present address: Department of Biology, University of New Mex- in accordance with 18 U.S.C. §1734 solely to indicate this fact. ico, Albuquerque, NM 87131. 9044 Downloaded by guest on September 30, 2021 Population Biology: Ellstrand et A Proc. Natl. Acad. Sci. USA 86 (1989) 9045 The plants making up the experimental populations were likelihood estimate of Y, the number of cryptic long-distance chosen so that the paternal contributions of each population matings. could be distinguished from those of the remaining popula- Furthermore, because pollen gametes from the natural tions. To attain this goal, each population was genetically populations can carry the same alleles at PGMI and PGM2 structured with different PGMI and PGM2 genotypes. Plants as the synthetic populations, such gametes might mimic gene making up population 1 had PGMI genotypes of22, 23, or 33 flow among the planted populations. Consequently, before with a PGM2 genotype of 33; those making up population 2 examining gene flow among the synthetic populations per se, had PGMI genotypes of 00, 01, or 11 with PGM2 genotypes a second analysis must be performed to estimate how many of 11, 12, 22, 13, 23, or 33; and those making up population pollen gametes from the natural populations would be mis- 3 had PGMI genotypes of22, 23, or 33 with PGM2 genotypes assigned as arising in alternate synthetic populations. 11, 12, or 22. In matings among these populations, paternal To illustrate our approach, consider population 1. This contributions from population 2 could be distinguished by a population had only PGMI alleles 2 and 3 and PGM2 allele progeny's PGMI genotype; the others are distinguished by 3. Gene flow from the other synthetic populations is detect- their PGM2 genotype. The PGMI allele frequencies of these able from progeny bearing PGMI alleles 0 or I or from loci in the natural populations were 0.211, 0.239, 0.385, and progeny bearing PGM2 alleles I or 2. Applying simple 0.165 for alleles 0, 1, 2, and 3, respectively; for PGM2, 0.220, probability rules, the number of undetected gametes from 0.392, and 0.388, respectively, for alleles 1, 2, and 3. These natural populations that mimic gene flow from populations 2 loci are also not linked (10). or 3 is given by: We transplanted the seedlings at the rosette stage. The three populations were planted in a triangular pattern, iso- T = (0) * [p(I) + p(2) - p(1) * p(2)] * (X + Y) lated from each other at distances of255, 280, and 400 m, with populations 2 and 3 the most distant and populations 1 and 3 where p(J) is the probability that a gamete from the natural the least distant. Each population consisted of 15 plants populations has allele 2 or 3 at PGMI and p(2) is the arranged in a 3 x 5 pattern with 2-m spacing. probability that a gamete from the natural populations has Flowers were naturally pollinated. Floral visitors (pre- allele 3 at PGM2. (The value of the expression within sumed pollinators) were mainly syrphid flies early in the brackets is the probability that the natural populations pro- season; numbers of honeybees and lepidoptera increased duce a gamete that would mimic gene flow from populations with the season as did the overall visitation rate. To test for 2 or 3.) temporal heterogeneity in gene flow, flowering stems on each Studies of Natural Populations. From 1982 to 1986, we used plant were labeled once a week. Fruits were harvested when paternity exclusion analysis to measure rates of apparent mature. About 150 to 200 seeds from each population were gene flow (11) for seven small, natural wild radish popula- collected from fruits fertilized during the weeks of April 11, tions in Moreno Valley and Riverside. The populations April 25, and May 9 (only 33 matured in population 3 for the studied were isolated from the nearest conspecific population third week). These seeds were germinated. Within 14 days of by distances ranging from 100 to 1000 m. Population sizes germination, the seedlings were electrophoretically analyzed were chosen to be similar, from 40 to 60 flowering individ- for the marker loci. uals. Gene flow rates were first estimated by comparing progeny genotypes with that of their maternal parent to obtain the paternal contribution (11). Paternal contributions that could RESULTS not be produced by the population from which the seeds were Synthetic Populations. Apparent gene flow. Apparent gene taken were the result of pollen gene flow ("apparent gene flow (either from other synthetic populations or from the flow," see ref. 12). Those contributions that could not be natural populations) varied considerably among our experi- assigned to any of the synthetic populations represent mat- mental populations and over time, ranging from as low as 0% ings from beyond the study site and were assigned to the for two sampling dates in population 3 to 5.4% for the first surrounding natural populations. Subtracting the gene flow sampling date in population 2 (Table 1). Some, but not all, of assigned to the natural populations from the apparent gene the interpopulational matings could be assigned directly to flow rate, we obtain the unassigned gene flow rate, that is, the natural populations; others could have been assigned to gene flow that could be assigned either to natural populations either the natural or the synthetic populations. We do not or to other synthetic populations. This method underesti- detail these assignment differences because our subsequent mates long-distance mating events because the natural pop- analysis (below) reveals that almost all of these progeny are ulations create some pollen gametes that have the same likely to have paternal parents in the relatively distant natural genotype as those in the experimental populations ("cryptic populations. The heterogeneity of apparent gene flow varied gene flow"; see ref. 12). significantly over populations and over time (Table 2). We were able to obtain a maximum likelihood estimate of Apparent plus cryptic gene flow. To refine our gene flow gene flow from the natural populations because all synthetic estimate, we lumped our data on apparent gene flow over populations were fixed for the same allele at the TPI and ACP time for each population. We then calculated our refined gene loci, and we had an estimate ofallele frequencies ofthose loci flow estimate by correcting for cryptic gene flow as detailed in the surrounding natural populations. To identify cryptic in Materials and Methods. We did not calculate a new gene flow, we start by defining the probability that the estimate for population 3 because the one progeny seedling surrounding populations will create a gamete indistinguish- resulting from gene flow was not detected from a rare allele able at TPI and ACP from those produced within the exper- ACP and TPI but from PGMI and PGM2. We conservatively imental populations. Because the surrounding populations leave this seedling as unassigned. The refined estimates for are large and in linkage equilibrium (N.C.E. and J. Lee, populations 1 and 2 (Table 1) are both about 66% greater than personal communication), this probability is simply the prod- those calculated directly. The populations are still heteroge- uct of the allele frequency at each locus, in this case, 0.951 neous for gene flow (y2 = 17.11; df = 2; P < 0.001). x 0.824 = 0.784. Call this probability 0; then apparent gene We then calculated the expected frequency of cryptic flow occurs with probability 1 - 6. Since we know the matings that will mimic those among synthetic populations, number ofapparent long-distance matings detected using rare -r. Because we cannot assign gene flow in population 3 (see alleles at ACP and TPI loci (call this number X), we can use above), we calculate T for only populations 1 and 2. In these the proportionality rule 6/(1 - 6) = Y/X to find the maximum populations, the value of X is the same as the frequency of Downloaded by guest on September 30, 2021 9046 Population Biology: Ellstrand et al. Proc. Natl. Acad. Sci. USA 86 (1989) Table 1. Gene flow by pollen into synthetic populations Table 3. Gene flow by pollen into natural populations of wild radish of wild radish Sampling date Gene flow, % Apparent Gene flow receipt (n) Apparent* Totalt Population Isolation, m gene flow, % Synthetic population 1 4/11/87 (221) 3.2 1982* 1000 8.6 4/25/87 (203) 2.5 1983* 600 8.2 5/9/87 (197) 1.5 1984* 100 17.9 Unassigned 1.3 0.0 1985At 200 9.8 Assigned to distant 1985Bt 200 18.0 natural populations 1.1 3.9 1986Af 150 4.0 Overall (621) 2.4 3.9 1986Bt 150 3.2 *Ref. 11. Synthetic population 2 4/11/87 (223) 5.4 tPreviously unpublished data. 4/25/87 (202) 2.0 tRef. 12. 5/9/87 (195) 0.5 Unassigned 1.7 0.0 may vary greatly. Kirkpatrick and Wilson (14) measured a Assigned to distant range of 0% to 15% gene flow from experimental plots of natural populations 1.0 4.5 cultivars of Cucurbita pepo to several single wild plants of Overall (620) 2.7 4.5 Cucurbita texana; the two species were isolated by over 450 m. Two paternity studies on natural stands of Pseudotsuga Synthetic population 3 4/11/87 (250) 0.0 menziesii give a picture of variation in gene flow by pollen in 4/25/87 (164) 0.6 this species. While Neale (15) found natural rates of inter- 5/9/87 (33) 0.0 - population mating varied from 20% to 27% at 161-m isolation. Unassigned 0.2 0.2 N. Wheeler and K. Jech (personal communication) measured Assigned to distant rates of4% to 15% at 2000 m. The combined data show almost natural populations 0.0 0.0 a 7-fold range in gene flow rate. Overall (447) 0.2 0.2 While many evolutionary biologists have assumed that *Fraction of interpopulation matings detected. gene flow rates are uniformly low within most species (e.g., tFraction of interpopulation matings detected plus those inferred ref. 16), these data challenge that view. The variation in gene statistically. Gene flow rates are too low to permit estimates for each flow rates observed for Raphanus, Cucurbita, and Pseudo- sampling date. tsuga suggest that gene flow rates are likely to be idiosyn- cratic for populations rather than characteristic of species. progeny originally unassigned (Table 1). Thus, essentially all In fact, we even found gene flow varied over time in our of the gene flow by pollen detected probably originated in the experimental populations, generally decreasing as the season more distant natural populations. progressed. The changing pollinator fauna with a correspond- Natural Populations. Simple paternity analysis revealed ing change in foraging behavior may account for this trend considerable and significant variation in gene flow by pollen (17). In contrast, a paternity study measuring interpopulation in our seven study populations (X2 test; df = 6; P < 0.001; mating in a small natural population ofwild radish (11) did not Table 3). Apparent gene flow did not show the predicted find a significant difference in long-distance paternity over monotonic decrease with increasing distance to the nearest the season. population despite the 10-fold range in such distances (Ken- Both ofour empirical studies showed that, at a spatial scale dall's X = 0.08; P >> 0.20; two-tailed test). In fact, the most of 100 to 1000 m, spatial isolation is not well correlated with isolated population had a gene flow rate in the middle (8.6%) our estimated gene flow rates. Our experimental study dem- of the range observed (3.2-18.0%). onstrated that gene flow occurring among small, proximal populations at distances of 255, 280, and 400 m was rare or nonexistent (<<1%). However, two of these populations DISCUSSION received moderate amounts of pollen gene flow from large, We used genetic markers in experimental and natural stands natural populations approximately twice as far as the distance of wild radish to measure the patterns of variation of gene to the nearest synthetic site. Apparently, the size of the flow by pollen into small populations. In both cases, we found pollen source populations plays a more important role than considerable heterogeneity among populations in their gene distance alone. If interplant mating distances are strongly flow receipt. Even though our experimental populations were leptokurtic, as has been previously reported for this species flowering at the same time and occurred within a few hundred (18, 19) and many other seed plants (20, 21), then the tail of meters of one another, we found a 22-fold range in gene flow such a distribution decreases very slowly over distance. estimates (0.2-4.5%). When we measured apparent gene flow Although the probability of an individual siring seed beyond into small natural populations in different years and loca- its own neighborhood is small, it is not zero. Coupling this tions, we found an almost 6-fold range (3.2-18.0%). nonzero probability with the large size of the surrounding Comparable data are rare for small populations of other natural populations (thousands of plants), makes the proba- natural plants but confirm that rates ofinterpopulation mating bility of some long-distance mating high. On the other hand, the small size of our experimental populations renders the Table 2. x2 contingency analysis of apparent gene flow probability of interpopulational mating among these popula- in wild radish* tions very low, even though the distances among the popu- Effect df X2 P lations are substantially less than the distances between the natural and experimental populations. Thus, the small per- Gene flow x population 2 13.29 0.0013 plant differences in the probability of mating based on Gene flow x sampling date 2 6.68 0.0354 distance are overwhelmed by the large numbers of pollen Gene flow x population x donors in the natural populations versus the small numbers in sampling date 4 5.16 0.2717 the synthetic sites. Interestingly, while the size of the pollen *Tests of partial association have been reported (see ref. 13). sink population has been thought to be an important deter- Downloaded by guest on September 30, 2021 Population Biology: Ellstrand et al. Proc. Natl. Acad. Sci. USA 86 (1989) 9047 minant of gene flow receipt (4), the importance of the size of twice as high (compare ref. 12). These gene flow rates are local pollen sources relative to the local pollen sink has been high enough and variable enough to have a major impact on rarely discussed (but see refs. 2 and 22). the evolution ofsmall populations ofwild radish. Conclusions Our gene flow estimates from seven natural populations about other outcrossing plant species await further study. also showed no decrease ofgene flow with isolation distance. In fact, the nonsignificant relationship of gene flow to dis- We thank Melissa Bartholomew, Janet Lee, and Lisa Patty for their contributions to the genetic analysis and Jeff Glazner for field tance was slightly positive. All of these populations were the assistance. We also appreciate the comments of three anonymous same size but varied with respect to a variety of ecological reviewers on an earlier draft ofthis article. This research was funded parameters, including primary pollen vectors, population by National Science Foundation Grants BSR-8219384 and BSR- shape and density, presence or absence of alternate hosts for 8505982 to N.C.E. the pollinators, plant biomass, and rates of flower produc- tion. The nearest congeners ofeach were always in relatively 1. Slatkin, M. (1985) Annu. Rev. Ecol. Syst. 16, 393-430. large populations, ranging from hjindreds to thousands of 2. Antonovics, J. (1976) Ann. Mo. Bot. Gard. 63, 224-247. 3. Wright, S. (1969) Evolution and Genetics ofPopulations (Univ. individuals. The various ecological differences among these of Oregon Press, Chicago), Vol. 2. populations vary so much that it would be impossible to 4. Handel, S. N. (1983) in Pollination Biology, ed. Real, L. determine which are the most important determinants ofgene (Academic, Orlando, FL), pp. 163-211. flow variation. Alternatively, because these descriptive stud- 5. Ellstrand, N. C. (1988) Trends Ecol. Evol. 3, S30-S32. ies measured only apparent rates of gene flow, it is possible 6. Schoen, D. J. & Stewart, S. C. (1987) Genetics 116, 141-157. that were obscured by large dif- 7. Squillace, A. E. & Long, E. M. (1981) in Pollen Management distance-dependent trends Handbook, ed. Franklin, E. C. (U.S. Department of Agricul- ferences in undetected cryptic gene flow. That explanation is ture, Washington, DC), Agricultural Handbook No. 587, pp. not a likely one, however, because in the few cases where 15-19. total gene flow rates have been calculated by our maximum 8. Munz, P. A. (1959) A California Flora (Univ. of California, likelihood method (above for our experimental populations Berkeley). and in ref. 12), the relative ranking of gene flow has not 9. Frost, H. B. (1923) Genetics 8, 116-153. changed. Experiments, like those detailed above, involving 10. Ellstrand, N. C. & Devlin, B. (1988) Am. J. Bot. 76, 40-46. 11. Ellstrand, N. C. & Marshall, D. L. (1985) Am. Nat. 126, populations of different size, shape, and dispersion, might 606-616. prove the best method for teasing apart the relative impor- 12. Devlin, B. & Ellstrand, N. C., Evolution, in press. tance of some of these factors. 13. Dixon, W. J. (1981) BMDP Statistical Software (Univ. of Most of the gene flow rates that we obtained could have California, Berkeley). important consequences for the small populations studied. In 14. Kirkpatrick, K. J. & Wilson, H. D. (1988) Am. J. Bot. 75, all but one of the populations studied, the gene flow rates 519-527. observed were of a magnitude sufficient to prevent differen- 15. Neale, D. B. (1983) Dissertation (Oregon State Univ., Corval- tiation by genetic drift caused by small population size (3). lis). And in some populations, we measured gene flow rates so 16. Ehrlich, P. R. & Raven, P. H. (1969) Science 165, 1228-1232. > 17. Schmitt, J. (1980) Evolution 34, 934-943. high that relatively high levels of natural selection (s 0.2) 18. Crane, M. B. & Mather, K. (1943) Ann. Appl. Biol. 30, 301-308. would be necessary to bring about local differentiation-(2). 19. Bateman, A. J. (1947) J. Genet. 48, 257-275. On the other hand, the great gene flow variation we observed 20. Levin,-D. A. & Kerster, H. W. (1974) Evol. Biol. 7, 139-220. could, in itself, be sufficient to promote differentiation (23). 21. Levin, D. A. (1981) Ann. Mo. Bot. Gard. 68, 233-253. Furthermore, our gene flow estimates in our descriptive 22. Grant, M. C. & Antonovics, J. (1978) Evolution 32, 822-838. studies are underestimates and the actual rates may be even 23. Levin, D. A. (1988) Am. Nat. 132, 643-651. Downloaded by guest on September 30, 2021