(=gerris) remigis (hemiptera

Evolution, 46(2), 1992, pp, 43D-444

GENETIC POPULATION STRUCTURE AND LEVELS OF GENE FLOW IN THE STREAM DWELLING WATERSTRIDER, AQUARIUS (=GERRIS) REMIGIS (HEMIPTERA: GERRIDAE) RICHARD F. PREzIOSI AND DAPHNE

J.

FAIRBAIRN

Department ofBiology. Concordia University, 1455 de Maisonneuve Blvd. West, Montreal, Quebec, H3G 1M8 CANADA Abstract. -Gene flow, in combination with selection and drift, determines levels of differentiation among local populations. In this study we estimate gene flow in a stream dwelling, flightless waterstrider, Aquarius remigis. Twenty-eight Aquarius remigis populations from Quebec, Ontario, New Brunswick, Iowa, North Carolina, and California were genetically characterized at 15 loci using starch gel electrophoresis. Sampling over two years was designed for a hierarchical analysis of population structure incorporating variation among sites within streams, streams within watersheds, watersheds within regions, and regions within North America. Hierarchical F statistics indicated that only sites within streams maintained enough gene flow to prevent differentiation through drift (Nm = 27.5). Above the level of sites within streams gene flow is highly restricted (Nm -s 0.5) and no correlation is found between genetic and geographic distances. This agrees well with direct estimates ofgene flow based on mark and recapture data, yielding an N, ofapproximately 170 individuals. Previous assignment ofsubspecific status to Californian A. remigis is not supported by genetic distances between those populations and other populations in North America. Previous suggestion of specific status for south-eastern A. remigis is supported by genetic distances between North Carolina populations and other populations in North America, and a high proportion of region specific alleles in the North Carolina populations. However, because of the high degree of morphological and genetic variability throughout the range ofthis species, the assignment ofspecific or subspecific status to parts of the range may be premature. Key words. -Allozymes, Aquarius, gene flow, genetic distance, genetic drift, migration, population structure, waterstrider.

Received December 19, 1990.

The amount of gene flow among local populations limits both the degree and the rate of differentiation of those populations. In the absence of selection, gene flow determines the degree to which population subunits may drift apart. When selection is present, gene flow may mitigate the effects of differential local selection (Steams and Sage, 1980). Theoretical and empirical studies are divided as to whether gene flow levels are high enough to maintain similarity among local populations (Mayr, 1963; Grant, 1980; Barbujani and Sokal, 1990) or alternatively, low enough to allow drift and selection to result in differentiation oflocal populations (Ehrlich and Raven, 1969; Slatkin, 1985a; Schnake-Green et aI., 1990; Devlin and Ellstrand, 1990; Arter, 1990; for a review see Slatkin, 1985b). Gene flow may be estimated by direct measurement of the migration of individuals among populations using mark and recapture techniques, or may be estimated indirectly based on some measure of the genetic similarity of populations. For some 430

Accepted July 16, 1991.

species, apparent dispersal ability, as assessed by observing movements ofindividuals, agrees well with estimates ofgene flow obtained by indirect methods (Murray and Clarke, 1984; Waples, 1987), while for other species these two types ofestimates disagree (Baker, 1981; Liebherr, 1986). Contradictory results obtained from direct and indirect measures of gene flow emphasize the importance of studies which estimate both measures. In this paper we present a study of the genetic population structure of the waterstrider, Aquarius (=Gerris) remigis (Andersen, 1990), and indirectly estimate levels of gene flow at the local and regional level. Direct estimates ofdispersal distances for A. remigis were provided by Fairbairn (1985a, 1985b, 1986). Waterstriders (Hemiptera: Gerridae) are semiaquatic true bugs that live on the water surface of ponds, pools, lakes or streams (Andersen, 1982). Population sites are discrete and well defined, and proportions of migrants can be predicted from the proportion of long-winged individuals (Vep-

POPULATION STRUCTURE OF AQUARIUS REMIGIS

sa1ainen, 1978). A. remigis is found on the surface of streams and is the most common and abundant waterstrider in North America (Drake and Harris, 1928; Polhemus and Chapman, 1979). In Canada, A. remigis are univo1tine or partially bivo1tine, overwinter as adults and reproduce in spring (Matthey, 1974; Galbraith and Fernando, 1977; Fairbairn, 1985a). In most populations, A. remigis are almost completely wingless with less than 1% winged individuals (Calabrese, 1979; for exceptions see Froeschner, 1962 and Fairbairn, 1985a). Both field and lab studies suggest that dispersal by flight is extremely rare (Fairbairn, 1986; Fairbairn and Desranleau, 1987). Bowdan (1978) examined locomotion in A. remigis and determined that, while A. remigis have become very coordinated at rowing, they have lost their coordination in walking and attempt to row even on hard dry surfaces. Thus, migration amongA. remigis populations occurs primarily along the water surface and, for the majority of A. remigis populations, can only occur between sites connected by water. Dispersal along a stream may occur by active migration of adults or passive downstream drift of adults and nymphs (Fairbairn, 1985a). Fairbairn (1985a, 1986) conducted an intensive mark and recapture study of movement of adult A. remigis in a watershed on Mont-St-Hilaire, Quebec (Canada). The greatest individual displacement occurred in spring and, even at this time, net displacement was unlikely to be more than 100 meters. Of 2,940 A. remigis marked and recaptured by Fairbairn (1986) only 2 individuals moved between streams, a distance of approximately 0.5 km. Since these streams flow into and out of a small lake, migration between streams was probably over the surface of the lake and not over land. Previous examination of the genetic population structure of A. remigis is confined to a single study. Zera (1981) compared the genetic population structure of A. remigis and the waterstrider Limnoporus canaliculatus in the eastern U.S. He concluded that A. remigis is a highly isolated, 'island' species, able to diverge genetically by selection or drift. However, no estimates of gene flow were given. In addition, none ofZera's (1981) sites were connected by wa-

431

ter and thus no information is available on genetic variation within a watershed for A. remigis. The purpose of this study is to assess the genetic population structure and estimate levels of gene flow in A. remigis on both local and regional scales. Based on Fairbairn's (1985a, 1985b, 1986) mark-recapture work, we expect genetic differentiation to occur within a stream if distances are large enough. Substantial genetic isolation among streams, even among streams in the same watershed, is expected based on the limited number of migrants found to move between streams (Fairbairn, 1986). Extreme isolation is expected among watersheds since dispersal by flight is so rare (Fairbairn, 1986; Fairbairn and Desranleau, 1987). Ifextreme isolation exists among watersheds then genetic divergence at this and higher levels may be independent ofgeographic distance. A. remigis is found throughout a large portion of North America and into Central America, and displays a high degree ofmorphological variability (Andersen, 1990). Andersen (1990) states that the extent of morphological variation of A. remigis throughout its range may warrant the assignment of specific or subspecific status for some regions. Divergence at at least the subspecific level has been proposed for A. remigis in both the southwest and southeast regions of North America (Michel, 1961; Calabrese, 1979). Analysis of genetic differentiation of A. remigis populations provides additional information on the taxonomic status of regional populations of A. remigis. MATERIALS AND METHODS

Sampling was designed to determine the degree of genetic differentiation at four levels of expected isolation: regions within North America, watersheds within regions, streams within watersheds, and sites within streams (Fig. 1, Table 1). Figure 2 illustrates the Quebec watersheds of Mont-St-Bruno and Mont-Tremblant. Sites within streams were sampled only in the third Quebec watershed, Mont-St-Hilaire (Fig. 3). The Ontario watershed sites are on streams that feed into opposite sides ofthe Thames river, near London, Ontario, at points approximately 28 km apart. The northern California sites are on streams that feed into op-

432

R. F. PREZIOSI AND D. J. FAIRBAIRN

FIG. 1. Map of North America showing locations of sites sampled in the study (black triangles). Site names can be found in Table I.

posite sides of the Klamath river at points approximately 25 km apart. A total of 28 sites were sampled and 13 sites were sampled in two consecutive years (Table 1). Sampling was conducted in the summer or fall when population sizes are the largest. Adult A. remigis were collected from streams using hand nets and kept alive on ice until returned to Montreal where they were placed

individually in 0.5 ml micro-centrifuge tubes and frozen at -60°C. Animals from California were frozen at -60°C in California and shipped to Montreal on dry ice. Whenever possible, 100 to 200 individuals were collected per site to increase the power of the chi-square tests of goodness-of-fit to Hardy-Weinberg expected genotypic distributions, used to test the underlying genetic

~

a

b T1

o

0.5 kin

1.0

o

2

3

kin

FIG. 2. Maps of the watersheds at (a) Mont-St-Bruno and (b) Mont-Tremblant showing locations of sample sites (black triangles).

433


models in the absence of breeding data (Fairbairn and Roff, 1980). Large sample sizes also increased the power of tests of independence of allele frequencies between sites and years. Horizontal starch gel electrophoresis was conducted using the 0.135 M tris-citrate buffer of Shaw and Prasad (1970) according to the methods of Zera (1981). The pH of the tris-citrate buffer was adjusted for each enzyme system (values follow E.C. numbers). The following 10 enzyme systems were used to detect 15 loci: Alkaline phosphatase (ALP-I, ALP-2, E.C. 3.1.3.1, pH = 7.3); Malate dehydrogenase (MDH-1, E.C. 1.1.1.37, pH = 7.3); Phosphogluconate dehydrogenase (PGD-1, E.C. 1.1.1.44, pH = 8.3); Glutamate-oxaloacetate transaminase (GOT-I, GOT-2, E.C. 2.6.1.1, pH = 7.3); Glycerol-3-phosphate dehydrogenase (GPD-l, E.C. 1.1.1.8, pH = 7.3); Glucose6-phosphate dehydrogenase (GDH-l, E.C. 1.1.1.49, pH = 7.3); Isocitrate dehydrogenase (lCD-I, ICD-2, E.C. 1.1.1.42, pH = 8.3); Lactate dehydrogenase (LDH-1, LDH2, E.C. 1.1.1.27, pH = 8.3); Malic enzyme (MEZ-1, MEZ-2, E.C. 1.1.1.40, pH = 7.3); Superoxide dismutase (SOD-I, E.C. 1.15.1.1, pH = 8.3). Gels were stained for enzymes according to Shaw and Prasad (1970) except for Malic enzyme gels which were stained according to Harrison (1977). Genotypic frequencies were tested for goodness-of-fit to Hardy-Weinberg equilibrium values following the methods ofSwofford and Selander (1981). Comparisons between observed and expected genotypic frequencies were made using chi-square tests or, where expected values were less than one, using Fisher's exact test. If more than two alleles were present at a locus, rare alleles were pooled for the analysis of genotypic frequencies. For sites where more than one locus was considered polymorphic (frequency of the most common allele less than 0.95), samples were tested for linkage disequilibrium using the LD79.FOR program of Weir (1990). For tests of linkage disequilibrium on sites with more than two polymorphic loci, significance levels were adjusted using the Bonferroni procedure (Weir, 1990). For sites sampled both years, allele frequencies were compared between years using a chi-square contingency test

H5

*

Q;J

H6

1987

()

() t) 1987

1988 ..-~

1988

H1

H2

e) 1987

0.0

0.5 km

~

~ 1987

~ 1988

1988

FIG. 3. Map ofMont-St-Hilaire watershed showing allele frequencies at the PGD-I locus from sites for 1987 and 1988. * P < 0.05.

(SYSTAT, Wilkinson, 1990). If allele frequencies were significantly different between years, Waples' (1989) modified chisquare program (TEMPTEST) was used to determine if changes in allele frequencies could be explained by a combination ofdrift and sampling error. Since effective population size (Ne ) affects the expected change due to drift, a minimum estimate of neighborhood size was calculated from data in Fairbairn (1985b, 1986). Data from both years were then pooled for further analysis (see Discussion). Mean heterozygosity and percent polymorphic loci were calculated for each site. Rogers' (1972) genetic distances were calculated for all pairs of sites and for all pairs in each level of the hierarchy outlined in Table I. Rogers' genetic distance was used in preference to Nei's genetic distance because both cluster analysis (Hillis and Moritz, 1990) and multidimensional scaling (Wilkinson, 1990) assume metricity of the

434

R. F. PREZIOSI AND D. J. FAIRBAIRN TABLE I. Region

Quebec

Details of hierarchical sampling design. Stream!

Watershed

Mont-St-Hilaire

South Creek

Site

HI H2

West Creek

H3 H4 H5

North Creek

H6

Mont-St-Bruno #1 Mont-St-Bruno #2

(Lac Du Moulin) (Lac Seigneurial)

BI B2

Mont-Tremblant

Chutes Croche

TI

Ruis. des Aulnes

T2

Ruis. Lac des Femmes

T3

Sample date

Size

08/87 08/88 08/87 08/88 08/87 08/88 08/87 08/88 08/87 08/88 08/87 08/88 09/87 09/87 08/88 09/87 08/88 09/87 08/88 09/87

157 151 201 184 119 170 167 110 66 160 124 116 38 207 157 179 241 226 213 200

07/87 08/88 08/88

109 68 70

07/87 07/88 07/87 07/88 07/87

29 28 98 28 35 168 165 75 377

Ontario

Thames River

Pottersburg Creek

Ll

Dingman Creek

L2

Iowa

Cedar Rapids # I

(Beaver Park)

CRI

Cedar Rapids #2

(Ellis Park)

CR2

Cedar Rapids #3

Squaw Creek

CR3

St. Andrew's #1

(Caithness)

NBI

St. Andrew's #2

Lelands Creek # I Lelands Creek #2

NB2 NB3

09/87 09/88 09/87 09/88

North Carolina

Deep River Yadkin River Blue Ridge

(Asheboro) (Tanglewood) (Dough ton Park)

NCI NC2 NC3

06/87 06/87 06/87

21 15 16

California

Klamath River

Beaver Creek Scott's River Prisoner's Creek San Justiano Creek Sespe Creek San Gabriel Creek

CI C2 C3 C4 C5 C6

09/88 09/88 09/88 09/88 11/88 11/88

47 133 142 100 100 100

New Brunswick

Santa Cruz Is. #1 Santa Cruz Is. #2 Los Padres Forest Angeles Forest

I Where stream names could not be determined a nearby town or landmark is given in parentheses.

distance measure. For contingency tests of allele frequencies among sites where more than one polymorphic locus was present, x2 values and degrees of freedom were combined to obtain an overall probability (Daniel, 1978 p. 339). Hierarchical F statistics were calculated for each of the levels in Table I. Gene flow was estimated from F values using the relationship, F S T = l/(4Nem + I). [For a comparison ofthis method with other methods of calculating gene flow see Slatkin and Barton (1989).] Cluster analysis was performed on a matrix of Rogers' ge-

netic distance between sites using the UPGMA clustering technique. All calculations for the above mentioned analyses, except where noted, were performed using the BIOSYS-l package of Swofford and Selander (1981). Matrices of Rogers' (1972) genetic distances and geographic distances were compared using a Mantel test (Mantel, 1967) following the methods of Sokal (1979) and Manly (1986 pp. 53-57). A plot of all pairs of genetic and geographic distances is also provided as suggested by Sokal (1979).


Multidimensional scaling (MDS) (Wilkinson, 1990) was used as an alternative method of examining the association between genetic and geographic distance. If an association exists between genetic and geographic distance, then the map produced by the multidimensional scaling procedure applied to genetic distance should resemble the geographic map. REsULTS

Banding patterns conformed to those expected based on enzyme structure (Richardson et al., 1986; Hillis and Moritz, 1990). Tests of deviation from Hardy-Weinberg equilibrium values were significant for 6 out of 94 comparisons. Of 94 comparisons, approximately 5 would be expected to be significant due to sampling error. Since no overall trend in heterozygote deficiency or excess was found for any locus, the basic genetic model ofinheritance for the systems analyzed cannot be rejected. None of the sites showed significant linkage disequilibrium, at a significance level of0.05, in either 1987 or 1988. Four out of 15 (26.7%) loci examined were monomorphic. Percentage of polymorphic loci by site ranged from 0.0% to 40.0%, and the mean heterozygosity by site ranged from 0.0 to 0.076. Of 53 alleles observed over all loci, 24 (45.3%) were unique to a region. In all there were 308 site-locus combinations (11 polymorphic loci x 28 sites). Of these, 263 (85.4%) were fixed for a single allele, 13 (4.2%) were fixed for an allele found only in that region, and 5 (1.6%) were fixed for an allele found only at that site. Percent polymorphic loci did not differ significantly among regions (F = 1.505; df= 4,21; P = 0.237; based on arcsine transformed data). Comparisons of allele frequencies between years showed significant differences for 8 of 26 comparisons. Allele frequencies at the PGD-llocus differed significantly for sites H5, B2, TI, and L1. Allele frequencies at the ALP-l locus differed significantly for sites H4, TI, and CR1. Site CRI showed significant differences in allele frequencies for the GPD-l locus. In some cases the changes in allele frequency were quite extreme. For the PGD-l locus at site LI the difference is due to a loss of three out of four alleles between years (x 2 = 53.4; df =

435

3). For the ALP-l locus at site CRI the difference is due to fixation of different alleles in the two years. Differences of this magnitude are unlikely to be produced by sampling error alone, given the relatively large sample sizes used. Waples' (1989) modified chi-square test determines if differences in allele frequencies between years can be explained by a combination of sampling error and genetic drift. However, the x2 statistic produced by Waples' test is sensitive to neighborhood size. A minimum estimate of neighborhood size in a linear habitat can be calculated from data in Fairbairn (1985a, 1985b) using the formula N; = 2 Duy;. (Wright, 1969 p. 303), where D is the density of individuals per unit distance and a is the standard deviation of movement of individuals along the habitat. Fairbairn (1985a) estimated a minimum density of 1.5 breeding individuals per meter for A. remigis. Matthey (1974) found a similar density of0.95 individuals per square meter in areas with surface current for streams in southern Alberta. Fairbairn (1985b) estimated the standard deviation of movement of overwintered individuals to be 32 meters. This does not include movement in the fall or over the winter and thus is a minimum estimate. The neighborhood size calculated from these estimates is 170 individuals or approximately 113 meters along a stream. Waples (1989) states that the modified chi-square test is applicable only to comparisons where significant differences have been found using a standard chi-square contingency test. Thus, Waples' (1989) modified chi-square was conducted for significant comparisons using a minimum estimate of N; of 170 individuals. All comparisons remained significant, indicating that the inter-annual differences can not be explained solely by the combined effects of sampling error and genetic drift assuming an N; of at least 170 individuals. Results of chi-square contingency tests of independence of allele frequencies among sites at all hierarchical levels indicate that significant differences exist among sites at all levels, even at the level of sites within a stream (all comparisons significant at P ::; 0.002). Differences among sites, even within a watershed, were large when compared to

436


.36 I

.32

28

.24

.20

.16

.12

.08

.04

.00

I

I

I

I

I

I

I

I

I Quebec,H1 Quebec,H2 Ontario, L1 Ontario, L2 New Brunswick, NB3 California, C4 california, C3

Quebec,H3 Quebec, H4

Quebec,H5 Quebec,H6

Quebec, T1 Quebec, T2 California, C1 California, C5 california, C6 california, C2

Iowa, CR1 lowa,CR2 lowa,CR3

Quebec, T3 Quebec,B1 Quebec, B2

New Brunswick, NB1 New Brunswick, NB2 North carollns, NC3

North Carolina, NC2 North Carolina, NC1

I .36

.32

28

.24

I .20

.16

.12

.08

.04

.00

Roger's Genetic Distance FIG. 4.

UPGMA dendrogram of Rogers' genetic distance among sites.

the previously described inter-annual differences at single sites. This is illustrated for the watershed at Mont-St-Hilaire in Figure 3. Although allele frequencies at the PGD1 locus differed significantly between years at site H5, this inter-annual variation is small when compared to the differences at this locus among streams. Cluster analysis of Rogers' genetic dis-

tances using a UPGMA algorithm produced the dendrogram shown in Figure 4. Since only 15 loci were examined, length of the branches is not assumed to be precise (Nei et al., 1985). However, the cophenetic correlation (0.92) indicates that the clustering pattern accurately reflects the underlying pattern of variation in the matrix ofgenetic distance (Sneath and Sokal, 1973 pp. 278-


437

TABLE 2. Average values for Rogers' (1972) genetic distance between sites with ranges given in brackets. Values on the diagonal are distances between sites within a region, values below the diagonal are distances between sites in different regions. Region

Quebec Ontario Iowa New Brunswick North Carolina California

Quebec

Ontario

Iowa

New Brunswick

North Carolina

California

0.076 (0.003-0.217) 0.060 (0.006--0.131) 0.123 (0.076--0.233) 0.139 (0.018-0.263) 0.360 (0.215-0.509) 0.094 (0.018-0.193)

0.016 (0.016--0.016) 0.087 (0.071-0.115) 0.100 (0.009-0.147) 0.347 (0.205-0.490) 0.052 (0.008-0.076)

0.029 (0.005-0.043) 0.165 (0.065-0.237) 0.324 (0.266--0.423) 0.079 (0.063-0.108)

0.163 (0.138-0.210) 0.353 (0.204--0.490) 0.135 (0.004--0.202)

0.240 (0.149-0.358) 0.357 (0.204--0.486)

0.040 (0.008-0.067)

279). This correspondence was confirmed by a Mantel test for a correlation between the original matrix of Rogers' genetic distance and the distance matrix reproduced from the cluster dendrogram (r = 0.934, P < 0.001). In some cases all sites within a region have clustered together [North Carolina (NCl, 2, 3), Iowa (CRl, 2, 3)]. However, in most cases sites within a region are not clustered together. For example, six sites from Quebec are clustered together, two are found in a cluster with two sites from Ontario, one site from New Brunswick, and two sites from California, and the remaining three sites from Quebec are not clearly clustered together or with any other group. Similarly, of the six California sites only four are clustered together, and sites from both Ontario and New Brunswick are found in at least two clusters. Even sites within single watersheds (HI-H6, Tl-T3, Ll-L2, NB2NB3, C l-C2) cluster more closely with sites from other watersheds, and often from other regions than with each other. This type of clustering indicates that the presence of particular alleles at a site is not necessarily associated with the region, or even the watershed, in which the site is located. The only clearly isolated cluster, and one which reflects geographic isolation, consists of the three North Carolina sites. The averaged estimates of Rogers' genetic distance among sites in different regions (Table 2) indicate that while the average genetic distance between sites within a re-

gion is generally smaller than the average genetic distance between sites in that region and sites in other regions, the ranges ofthese genetic distances greatly overlap. Overall, the highest levels of genetic divergence are between North Carolina sites and all other sites. Because the North Carolina sites stand out as divergent from all other sites, subsequent analyses were conducted both with and without the North Carolina sites. F statistics and gene flow estimates (Table 3) show the same pattern of site differentiation as found in the matrices of averaged genetic distances. The pattern of differentiation revealed by the F values remains similar when the North Carolina sites are removed from the analysis. Because of the method of calculation, all F values are affected by the overall reduction in variance (Wright, 1978 p. 89). As expected, the F values for regions within total, watersheds within total and watersheds within regions are reduced. Reductions in the latter two F values were expected because of the high degree of differentiation among watersheds within North Carolina (Table 2, Fig. 4). Both analyses (with and without North Carolina) indicate very little genetic differentiation within a stream. Above the level of sites within a stream, subgroups are highly divergent, with the highest divergence occurring among sites within the total and streams within the total. Because the removal ofthe North Carolina sites does not alter the pattern ofdivergence revealed by the F values,

438

R. F. PREZIOSI AND D. J. FAIRBAIRN TABLE 3. Hierarchical F statistics and gene flow for all sampling levels. N.C. sites included Subgroup

Total group

Site Site Stream Stream Watershed Watershed Region a Nm calculated from FST

Stream Total Watershed Total Region Total Total ~

1/(4Nem

N.C. sites removed

FST

Nm"

FST

Nm"

0.009 0.766 0.463 0.764 0.340 0.560 0.333

27.53 0.08 0.29 0.08 0.49 0.20 0.50

0.011 0.694 0.494 0.690 0.263 0.388 0.169

22.48 0.11 0.26 0.11 0.70 0.39 1.29

+ I).

subsequent quantitative analyses of this pattern is presented only for the full data set. Each of the Fstatistics represents the proportion of allelic variance within the group that is due to variance among the subgroups. The proportion of variance within the subgroups within the group is 1 - F. Thus 0.9% of the allelic variance within streams is found among sites and 99.1 % is found within the sites. Similarly, 46.3% of the allelic variance within watersheds is found among streams and 53.7% within streams. F statistics may also be interpreted as the proportion of the total variance, or gene diversity, found at each level ofthe hierarchy. For the lowest level, sites, the proportion of total variance is calculated as 1 - F (sites within total). Thus, 23.4% of the total variance is found within sites. For the highest level, among regions, the proportion oftotal variance is simply F (regions within total) which is 33.3%. For intermediate levels the proportion of total variance is calculated as the difference between the Fvalues, relative to the total, for the level in question and the level above the level in question. Thus 0.2% of the total variance is found among sites, 20.4% among streams, and 22.7% among watersheds. These results indicate that the proportion of total variance is similar at the three highest levels. The low proportion of total variance at the level of among sites shows that there is very little genetic isolation among sites when these sites are considered in the context of streams that are genetically very isolated. Estimates of gene flow among sites are based on an island model of population structure. For A. remigis a more appropriate model might be a two dimensional step-

ping-stone model. Estimating gene flow from the two dimensional stepping-stone model requires an estimate ofthe ratio of mutation to migration rates (Slatkin, 1985b). Since we cannot accurately estimate this ratio, and since Slatkin (l985b) indicates that gene flow estimates from the two models are qualitatively similar, we have used the simpler island model. Gene flow estimates (Table 3) indicate that only sites within streams exchange enough individuals to avoid substantial genetic differentiation through genetic drift (Nm > 1). Gene flow among sites within streams was approximately two orders of magnitude larger than gene flow between streams within a watershed. The increase in gene flow from the level ofstreams within watersheds (Nm = 0.29) to the level of watersheds within regions (Nm = 0.49) or regions within North America (Nm = 0.50), may be explained by the increasing number of sites sampled at higher levels in the hierarchy. Comparisons at the level of watersheds or regions are based on pooled allele frequencies and thus include most or all of the major alleles, resulting in an apparent decrease in the level of differentiation and an apparent increase in the level of gene flow. The Mantel test for a correlation between matrices of Rogers' (1972) genetic distance and geographic distance was not significant (r = -0.035, P = 0.55), indicating that genetic similarity does not correspond to geographic proximity. Similar results were obtained when the North Carolina sites were removed (r = 0.104, P = 0.17). The plot of all pairs ofgenetic and geographic distances shows the lack of association between these measures and demonstrates that the lack of association is not dependent on the inclusion ofthe North Carolina sites (Fig. 5). An

439

POPULATION STRUCTURE OF AQUARIUS REMIGIS 0.8

[!]

• ..•u C

-

'6

.•

.!!

Q t.

0.5

0.4

t.

N

D

-0-...

0.2

IIlI 0

0.1

II:

~ D

'"

~

'"

t.

~ 0.0 0

'l,

*~

t. ~

D c5'D

D D D

8

D

N N

·2

,

D

i

D

a

B DD

c~

/At.

l!l

t.

D

~

Q

.,

i

t.

0.3

GI

B

t. t.

...

B

1\

~

C

IIlI

i

t.

III D

...

.-.-3

qj

D

8

,. • • ~

-!o C

~B

8

1000

2000

Ii 3000

4000

-4

.,

0.5

-0.5

2

Q

CD

~ 11.5 0

~

8

N

~

'iii

5000

geographic: dlatanc:e (km)

Q

0.5

FIG. 5. Plot of Rogers' (1972) genetic distance against geographic distance for all pairs of sites. Triangles indicate comparisons including North Carolina sites; squares indicate comparisons not including North Carolina sites.

B

-0.5

Q

., -'.5

-as

alternative method of visually exammmg the relation between genetic and geographic distance is multidimensional scaling (MDS). Figure 6a shows the map produced by the multidimensional scaling procedure applied to the matrix of genetic distances between watersheds; it is clear that the MDS map produced from Rogers' (1972) genetic distances bears no resemblance to a geographic map of the watersheds sampled. MDS analysis excluding the North Carolina sites (Fig. 6b) only expands the scale of dimension 2 and has little effect on the overall pattern.

Q

B

-2

·'.5

.,

-0.5

0.5

Dimension (1) FIG. 6. Map produced by MDS analysis of Rogers' genetic distance both (a) including and (b) excluding the North Carolina watersheds (see text). Symbols represent watersheds and are coded by region (Q = Quebec, 0 = Ontario, I = Iowa, B = New Brunswick, C = California, N = North Carolina).

Estimates ofgene flow and population differentiation based on F S T assume an equilibrium has been reached between migration and drift. The number of generations taken to reach equilibrium for an island model is of the same order of magnitude as 11m(Slatkin, 1985b). Aquarius remigis have DISCUSSION expanded their range since the Pleistocene In this study we show that gene flow may (over the last 10,000 years). Because of the be highly restricted among local popula- low levels of gene flow at higher levels (wations. For Aquarius remigis, the population tersheds, regions), differentiation is expectsubdivision level at which gene flow may ed due to drift, and the degree of differenprevent differentiation ofsubdivisions is that tiation will increase until equilibrium is of sites within streams. All levels of sub- reached. If populations have not reached division above sites within streams (i.e., equilibrium, our F values will underestistreams within watersheds, watersheds mate the degree of differentiation expected within regions, regions within North Amer- at equilibrium and our estimates of gene ica) have greatly reduced gene flow and may flow (Nm) will overestimate the true levels. differentiate from each other through the Estimates ofgene flow also assume that popeffects of drift. Each aspect of the effect of ulation structure is fixed and that gene flow population subdivision is examined in de- occurs through migration of individuals among sites. IfIocal populations are subject tail below.

440


to extinction and recolonization then, for the island model, the estimated gene flow actually represents N(m + el2), where e is the probability ofa population going extinct each generation (Slatkin, 1985a). The effects of extinction and recolonization thus lead to overestimates of gene flow. For A. remigis, these effects may be significant both within and among streams. Thus, our estimates of gene flow are more likely to overestimate than to underestimate true levels.

cies differed significantly among sites within a stream. Since sample sites were more than 113 meters apart (actual distances were around 250 m) these results agree well with the neighborhood size estimated from Fairbairn's (1985a, 1985b) data.

Comparison ofAllele Frequencies between Years Comparisons of allele frequencies between years were significant for 8 out of 26 comparisons at six sites. These comparisons Neighborhood Size and Genetic remain significant even when analyzed usVariation within Watersheds ing Waples' (1989) test which corrects for Variation among sites within streams was both sampling error and genetic drift. Since measured only within the watershed at Waples' test adjusts for changes due to geMont-St-Hilaire. Generalizations about netic drift it is highly sensitive to the estivariability at this level must be made with mate of N; used. We have used an estimate caution. However, the streams within this of 170 based on movements detected watershed are typical of A. remigis habitat through mark and recapture. Waples' analthroughout North America (Fairbairn, ysis indicates that the combination of ge1985a, 1985b), and therefore we have no netic drift and sampling error could account reason to suppose that levels of genetic dif- for the observed inter-annual differences ferentiation within the streams at Mont-St- only if N; was less than halfofour estimated Hilaire are atypical. Although allele fre- value. Considering the high overwinter quencies differ significantly among sites mortality of A. remigis (76% to 89%; Matwithin a stream, most of the genetic vari- they, 1974; 60% to 95%; Fairbairn, 1986), ability is found at higher levels. The hier- the founder effects that would result from archical analysis indicates that the greatest limited dispersal, and the fact that these efdifferentiation occurs at the level of streams fects may be compounded several times, it within watersheds. The levels of differen- is possible that neighborhood sizes are much tiation among watersheds within regions and smaller than can be estimated from markamong regions within the total were similar recapture data. If so, genetic drift may acand slightly lower than differentiation among count for these inter-annual differences. streams within a watershed. These results Waples' test assumes discrete generations support the initial hypothesis of high levels and no migration. The former assumption ofgenetic isolation among streams within a holds for univoltine populations of A. rewatershed. The largest level of genetic di- migis. Southern populations may be parvergence was expected to be found at the tially or fully bivoltine with overlap of genlevel of watersheds, but was found at the erations in the summer months. Ofthe sites even lower level of streams. sampled for two years, only the Iowa samGene flow estimates calculated from these ples (CR 1, CR2) may contain mixed genF statistics indicate that only sites within a erations. The assumption of no migration stream exchange a sufficient number of in- is virtually true at and above the level of dividuals to prevent genetic differentiation streams. However, we have established that through the effects ofdrift. This sharp cutoff migration occurs among sites within streams in gene flow above the level of sites within and that these sites may have significantly streams indicates that the neighborhood size different allele frequencies. Migration among of A. remigis is at the level of a stream or sites within streams may therefore produce smaller. Isolation by distance was expected inter-annual changes in gene frequencies within a stream only if streams were long within sites, over and above those expected enough to include more than one neighbor- due to drift alone. hood. For the Mont-St-Hilaire streams this Variation between years could also be appears to be the case since allele frequen- produced by selection. For the Mont-St-Hi-


laire watershed the ALP-110cus showed significant inter-annual differences at site H4, and the PGD-Ilocus showed significant inter-annual differences at site H5. If either difference were due to selection then we might expect similar differences at the other sites within the Mont-St-Hilaire watershed and particularly within the same streams. Such similarity was not observed. The significant difference at the ALP-I locus was due to a decline in the less common allele. Over the same time period this allele increased in frequency, but not significantly, at the other five sites. The significant interannual change in allele frequency at the PGD-I locus was due to a decline in the frequencies of both the rarest and most common alleles. Ofthe other two sites within the same stream, one showed an increase in the frequency of the rare allele and the other showed no change. The common allele did not change in frequency at either site (Fig. 3). A fourth explanation for apparent changes in allele frequencies between years is the sampling of sibling groups. This problem has been reported in several studies (Varvio-Aho, 1979; Parkinson, 1984; Guttman and Weigt, 1989; Reisenbichler and Phelps, 1989). Fairbairn (1985b) estimated that the mean net movement of summer born adult A. remigis before overwintering was only 7.2 m, compared with a distance of39.5 m for A. remigis in the spring after overwintering. Thus, sibling groups may remain together until late fall when they leave the water surface to diapause. Lack of mixing of sibling groups may be especially true in small streams where the surface area may be reduced during the summer because of drying. If individuals do not disperse along the stream, samples of individuals taken in the late summer or fall may consist of a small number of sibling groups. Samples with this type of error could be expected to show differences in allele frequencies between years since it is a small number of sibling groups and not really the neighborhood that has been sampled. If this type of sampling error is present, allele and genotype frequencies from a single sample may not accurately reflect frequencies of the neighborhood. Therefore, where two years of data were available for a site, the data

441

were combined for all other analyses on the assumption that the combined data, being representative ofmore sib-groups, would be more representative ofthe neighborhood allele frequencies. If a problem of sampling sib-groups exists, care must be taken that observed differences among sites are not due to this sampling problem. Figure 3 illustrates that even where significant differences exist between years (site H5) these differences are much smaller than differences among sites. Additionally, the differences among sites within a stream are smaller than differences among sites from separate streams (Fig. 3). If sampling ofsib-groups was creating a bias in our estimates of allele frequencies, then differences between years would be expected to be similar in magnitude to differences among sites. The only case in which sampling of sib-groups could have produced a pattern such as found in Figure 3 is when differences among sib-groups within a stream are small relative to differences among sib-groups among streams. This case however, leads to the same conclusions concerning degrees of isolation as are reached when sampling of sib-groups is ignored. M .acrogeographic Variation In a comparison of genetic population structure in the eastern U.S. between the long-winged L. canaliculatus and the almost wingless A. remigis, Zera (1981) found that A. remigis populations were highly genetically divergent and suggested that this was due to a lack of gene flow among populations, small population size, bottlenecks, and founder effects. Our results also show high levels of genetic isolation and differentiation for all levels above sites within streams. At none of these levels is the estimated gene flow high enough to prevent genetic differentiation by drift. Comparison of these results with those reported for other species within the Gerridae (Varvio-Aho, 1979; Varvio-Aho and Pamilo, 1979; Zera, 1981) suggests a negative relationship between geographic isolation and the dispersal capacity of the species as assessed by percent long-winged individuals. Varvio-Aho (1979) reported a GST of 0.055 for populations of the long-winged species, G. odontogaster. Genetic differentiation among populations

442


of wing dimorphic species ranges from F ST = 0.082 for Limnoporus canaliculatus (calculated in McCauley and Eanes, 1987 from data in Zera, 1981) to GST = 0.283 for Gerris lacustris (Varvio-Aho and Pamilo, 1979, 1981). All of these values are small when compared to the F ST for values calculated for A. remigis at the level of either sites within watersheds (F = 0.500) or streams within regions (F = 0.627) (the most nearly comparable hierarchical F values). These data indicate that among Gerridae species, genetic divergence of population subunits is higher for species with decreased dispersal ability (but see Varvio-Aho and Pamilo, 1982 for possible confounding effects of population size in marginal habitat or at the edge of a species range). In continuous populations, genetic drift should lead to a correlation between geographic and genetic distance among population subunits. This would also be true for subdivided populations if the gene flow among subunits is proportional to distance. Ifno such association exists between genetic and geographic distance selection may be acting independently on different subunits (overriding the effects of any gene flow that may exist) or there may be very low or no gene flow between neighboring subunits (making the neighborhood size equal to or less than the subunit examined). For selection to explain lack of correlation between genetic and geographic distance would require that it is acting in a different manner in different population subunits and that patterns of selection are themselves uncorrelated with geographic distance. The lack of a significant association between genetic and geographic distances for A. remigis reflects the high degree of genetic isolation at all levels above sites within streams. A similar lack of association between genetic and geographic distances has been found for other 'island' insect species (King, 1987; Guttman and Weigt, 1989). The matrix of genetic distances averaged by region indicates a broad overlap of distances within regions and distances among regions. Approximately 85% ofall site-locus combinations were fixed for a single allele and approximately 45% of all alleles were unique to a region. This high level of fixation of alternate alleles without geographic

pattern indicates the effects of drift. Cluster analysis of Rogers' (1972) genetic distances grouped some sites by region but the general pattern of clustering of sites appears to be random. Our results at this level agree well with Zera's 1981 conclusions that A. remigis populations are highly genetically isolated and divergent due to the effects of drift. Following the Pleistocene, aquatic habitat was abundant in North America. Since then there has been a general warming and drying resulting in a loss of aquatic habitat. The pattern of genetic differentiation observed for A. remigis is probably due to the random loss of alleles through local extinction and recolonization accompanying loss of habitat. Andersen (1990) suggested that the high degree of morphological variability throughout the range of A. remigis may justify the assignment of specific or subspecific status to some portions ofthe species range. Calabrese (1979) has proposed that western A. remigis in California and Oregon are a separate subspecies on the basis of morphology and proportion of long-winged individuals. The average genetic distances between regions indicate that the California region is not more different from Quebec, Ontario, Iowa and New Brunswick than these regions are from each other. Aquarius remigis shows strong regional differentiation throughout its range, and the degree of differentiation of California populations is not particularly striking in this context. Genetic distances among all of these regions would, in fact, be consistent with assignment of subspecific status to each region (Ayala et al., 1974; Nei, 1976). North Carolina has genetic distances from other regions which would suggest divergence at the species level (op. cit.) as suggested by Michel (1961) on the basis ofmorphological data of male genitalia. The North Carolina region had high variability between sites and sample sizes were small, yet out of28 alleles found in North Carolina populations, II were unique to that region. In conclusion, A. remigis populations are highly subdivided. Genetic differentiation is highest at the level of streams and occurs within the context oflower levels ofgenetic differentiation among watersheds and regions. Gene flow is rare among streams


within a watershed and only sites within a stream maintain high enough levels of gene flow to prevent genetic differentiation by drift. The mark and recapture data suggest that neighborhood size is smaller than the population of a single stream. The possibility of isolation by distance within a stream is supported by significant differences in allele frequencies among sites within streams. The overall pattern of genetic variation in A. remigis suggests that drift, bottlenecks and founder events playa large role in defining levels of population differentiation. These factors occur within the context of a lack of gene flow due to winglessness for eastern populations. The genetic population structure of A. remigis conforms to the views of Ehrlich and Raven (1969) that gene flow is highly reduced between populations and that local populations, in this case streams or possibly subunits of streams, are the units of evolutionary importance. ACKNOWLEDGMENTS

We are most grateful to R. B. Selander, R. S. Waples and B. S. Weir for kindly providing copies of their computer programs. We also thank D. Rofffor assistance in collecting samples from North Carolina and California. The manuscript has benefited from the critical comments of D. Green, L-A. Giraldeau, D. Good, E. Maly, D. Roff, K. Sittmann, R. Waples and two anonymous reviewers. We are grateful for their help. This work was supported by NSERC grant A0347 to D.J.F. LITERATURE CITED

ANDERSEN, N. M. 1982. The Semiaquatic Bugs (Hemiptera, Gerromorpha). Phylogeny, Adaptations, Biogeography, and Classification. Entomonograph Vol. 3, Scandinavian Science Press Ltd., KJampenborg, Denmark. - - . 1990. Phylogeny and taxonomy of water striders, genus Aquarius Schellenberg (Insecta, Hemiptera, Gerridae), with a new species from Australia. Steenstrupia 16:37-81. ARTER, H. E. 1990. Spatial relationship and gene flow paths between populations of the alpine snail Arianta arbustorum (Pulmonata: He1icidae). Evolution 44:966-980. AYALA, F. J., MARTIN L. TRACEY, DENNIS HEDGECOCK, AND ROLLIN C. RICHMOND. 1974. Genetic differentiation during the speciation process in Drosophila. Evolution 28:576-592. BAKER, A. E. M. 1981. Gene flow in house mice:

443

Introduction of a new allele into free-living populations. Evolution 35:243-258. BARBUJANI, G., AND R. R. SoKAL. 1990. Zones of sharp genetic change in Europe are also linguistic boundries. Proc. Natl. Acad. Sci. 87:1816-1819. BoWOAN, E. 1978. Walking and rowing in the water strider, Gerris remigis: I. A cinematographic analysis of walking. J. Compo Physiol. 123:43-49. CALABRESE, D. 1979. Population and subspecific variation in Gerris remigis Say. Ent. News 85:2728. DANIEL, W. W. 1978. Biostatistics: A Foundation for Analysis in Health Sciences, 2nd ed. John Wiley and Sons, N.Y., USA. DEVLIN, B., AND N. C. ELLSTRAND. 1990. The development and application of a refined method for estimating gene flow from Angiosperm paternity analysis. Evolution 44:248-259. DRAKE, C. J., AND H. M. HARRIS. 1928. Concerning some North American water-striders with descriptions of three new species. Ohio J. Sci. 28:269-275. EHRLICH, P. R., AND P. H. RAVEN. 1969. Differentiation of populations. Science 163: 1228-1232. FAIRBAIRN, D. J. 1985a. ComparativeecologyofGerris remigis (Hemiptera: Gerridae) in two habitats: A paradox of habitat choice. Can. J. Zool. 63:25942603. - - . 1985b. A test of the hypothesis of compensatory upstream dispersal using a stream-dwelling waterstrider, Gerris remigis Say. Oecologia 66: 147153. - - . 1986. Does alary polymorphism imply dispersal polymorphism in the waterstrider, Gerris remigis? Ecol. Entomol. 11:355-368. FAIRBAIRN, D. J., AND L. DESRANLEAU. 1987. Flight threshold, wing muscle histolysis, and alary polymorphism: Correlated traits for dispersal in the Gerridae. Ecol. Entomol. 12:13-24. FAIRBAIRN, D. J., AND D. A. ROFF. 1980. Testing genetic models of isozyme variability without breeding data: Can we depend on the chi-square? Can. J. Fish. Aquat, Sci. 37:1149-1159. FROESCHNER, R. C. 1962. Contributions to a synopsis of the Hemiptera of Missouri, Part V. Am. MidI. Nat. 67:208-240. GALBRAITH, D. F., AND C. H. FERNANDO. 1977. The life history of Gerris remigis (Heteroptera: Gerridae) in a small stream in southern Ontario. Can. Entomol. 109:221-228. GRANT, V. 1980. Gene flow and the homogeneity of species populations. BioI. Zbl. 99:157-169. GurrMAN, S. I., AND L. A. WEIOT. 1989. Macrogeographic genetic variation in the Enchenopa binotata Complex (Homoptera: Membracidae). Ann. Entomol. Soc. Am. 82:156-165. HARRISON, R. G. 1977. Patterns of variation and genetic differentiation in closely related species: The field crickets ofeastern North America. Ph.D. Diss. Cornell University, N.Y., USA. HILLIS, D. M., AND C. MORITZ. 1990. Molecular Systematics. Sinauer Associates, Inc. Sunderland, MA USA. KINo, P. 1987. Macro- and microgeographic structure of a spatially subdivided beetle species in nature. Evolution 41:401-416. LIEBHERR, J. K. 1986. Comparison of genetic varia-

444


tion in two Carabid beetles (Coleoptera) ofdiffering vagility. Ann. Entomol. Soc. Am. 79:424-433. MANLY, B. J. F. 1986. Multivariate Statistical Methods: A Primer. Chapman and Hall, N.Y., USA. MANTEL, N. 1967. The detection ofdisease clustering and a generalized regression approach. Cancer Res. 27:209-220. MATTHEY, W. 1974. Contributional'ecologiedeGerris rernigis Say sur deux etangs des Montagnes Rocheuses. Bull. de la Soc. Entomol. Suisse 47:85-95. MAYR, E. 1963. Populations, Species, and Evolution. Harvard University Press. Cambridge, MA USA. McCAULEY, D. E., AND W. F. EANES. 1987. Hierarchical population structure analysis of the milkweed beetle, Tetraopes tetraophthalmus (Forster). Heredity 58: 193-20 I. MICHEL, F. A. 1961. The taxonomic value ofthe male genitalia of the genus Gerris Fabricus (Hemiptera: Heteroptera: Gerridae). M.Sc. Thesis. Oregon State University, OR USA. MURRAY, J., AND B. CLARKE. 1984. Movement and gene flow in Partula taeniata. Malacologia 25:343348. NEI, M. 1976. Mathematical models of speciation and genetic distance, Pp. 723-765. In S. Karlin and E. Nevo (eds.), Population Genetics and Ecology. Academic Press, N.Y., USA. NEI,M., J. C. STEPHENS, AND N. SAITOU. 1985. Methods for computing the standard errors of branching points in an evolutionary tree and their application to molecular data from humans and apes. Mol. Biol. Evol. 2:66-85. PARKINSON, E. A. 1984. Genetic variation in populations of steelhead trout (Salrna gairdnerii in British Columbia. Can. J. Fish. Aquat. Sci. 41:14121420. POLHEMUS, J. T., AND H. C. CHAPMAN. 1979. Family Gerridae-Water Striders, Pond Skaters, Wherrymen. In A. S. Menke (ed.), The Semiaquatic and Aquatic Hemiptera of California. Bull. California Insect Surv. 21. RICHARDSON, B. J., P. R. BAVERSTOCK, AND M. ADAMS. 1986. Allozyme Electrophoresis. Academic Press, N.Y., USA. REISENBICHLER, R. R., AND S. R. PHELPS. 1989. Genetic variation in steelhead (Salrna gairdnerii from the north coast ofWashington. Can. J. Fish. Aquat, Sci. 46:66-73. ROGERS, J. S. 1972. Measures of genetic similarity and genetic distance. Studies in Genet. VII. Univ. Texas Publ. 7213:145-153. SCHNAKE-GREEN, J. E., L. W. ROBBINS, AND D. K. TOLLIVER. 1990. A comparison of genetic differentiation among populations of two species of mice (Perarnyscus). Southwest Nat. 35:54-60. SHAW, C. R., AND R. PRAsAD. 1970. Starch gel electrophoresis of enzymes- A compilation of recipes. Biochem. Genet. 4:297-320.

SLATKIN, M. 1985a. Rare alleles as indicators ofgene flow. Evolution 39:53-65. - - - . 1985b. Gene flow in natural populations. Annu. Rev. Ecol. Syst. 16:393-430. SLATKIN, M., AND N. H. BARTON. 1989. A comparison of three indirect methods for estimating average levels of gene flow. Evolution 43: 1349-1368. SNEATH, P. H. A., AND R. R. SaKAL. 1973. Numerical Taxonomy. W. H. Freeman and Co., San Francisco, CA USA. SOKAL, R. R. 1979. Testing statistical significance of geographic variation patterns. Syst. Zool. 28:227232. STEARNS, S. c., AND R. D. SAGE. 1980. Maladaptation in a marginal population of mosquito fish, Garnbusia affinis. Evolution 34:65-75. SWOFFORD, D. L., AND R. B. SELANDER. 1981. BlOSYS-I: A FORTRAN program for the comprehensive analysis of electrophoretic data in population genetics and systematics. J. Hered. 72:281-283. VARVIO-AHO, S. 1979. Genic differentiation ofGerris odontogaster populations. Hereditas 91:207-214. VARVIO-AHO, S., AND P. PAMIw. 1979. Genic differentiation of Gerris lacustris populations. Hereditas 90:237-249. - - - . 1981. Spatio-temporal microdifferentiation ofwater-strider (Gerris) populations. Entomol. Res. Camb. 37:253-263. - - - . 1982. Genic differentiation ofnorthern Finnish water-strider (Gerris) populations. Hereditas 92: 363-371. VEPSALAINEN, K. 1978. Wing dimorphism and diapause in Gerris: Determination and adaptive significance, pp. 218-253. In H. Dingle (ed.), Evolution of Insect Migration and Diapause. Springer, N.Y., USA. WAPLES, R. S. 1987. A multispecies approach to the analysis of gene flow in marine shore fishes. Evolution 41:385-400. - - - . 1989. Temporal variation in allele frequencies: Testing the right hypothesis. Evolution 43: 1236-1251. WEIR, B. S. 1990. Genetic Data Analysis. Sinauer Associates, Sunderland, MA USA. WILKINSON, L. 1990. SYSTAT: The System for Statistics. Evanston, IL USA. WRIGHT, S. 1969. Evolution of the Genetics of Populations: 2. The Theory of Gene Frequencies. University of Chicago Press, Chicago, IL USA. - - - 1978. Evolution and the Genetics of Populations: 4. Variability in and Among Natural Populations. University of Chicago Press, Chicago, IL USA. ZERA, A. J. 1981. Genetic structure of two species of waterstriders (Gerridae: Hemiptera) with differing degrees of winglessness. Evolution 35:218-225. Corresponding Editor: J. Arnold