Oecologia (1996) 106:308 316

© Springer-Verlag 1996

M i k a e l L 6 n n • H o n o r C. Prentice • Karin B e n g t s s o n

Genetic structure, allozyme.habitat associations and reproductive fitness in Gypsophila fastigiata ( Caryophyllaceae )

Received: 20 June 1995 / Accepted: 18 November 1995

Relationships between allozyme differentiation, habitat variation and individual reproductive success were examined in local populations of a perennial herb, Gypsophila fastigiata, on the Baltic island of Oland (Sweden). Relatively little (c. 2%) of the total allozyme diversity in this largely outcrossing species is explained by differentiation between sites tens of kilometres apart. The low level of geographic differentiation suggests that gene flow between sites is, or has recently been, extensive. Yet the component of allozyme diversity due to differentiation between plots (only tens of meters apart) within sites is 3 times larger than the between-site component of diversity. Allozyme variation, especially at the Pgi-2 locus, is significantly associated with habitat variation within sites. Different allele x habitat combinations for the Pgi-2 locus are associated with differences in individual reproductive fitness. Differential selection in different local habitats may thus contribute to the fine-scale structuring of genetic diversity within sites. Abstract

Key words Allozyme-habitat associations • Genetic structure. Gypsophila fastigiata " Reproductive fitness

Introduction Genetic variation in plants is patchily distributed on different geographic scales within and between popuM. L6nn ([~)1 . H, C. Prentice2 ' K. Bengtsson Department of Ecological Botany, Uppsala University, Villaviigen 14, S-752 36 Uppsala, Sweden FAX +46(0)18 553419 Present address:

1Department of Genetics, Uppsala University, Box 7003, S-750 07, Sweden 2Department of Systematic Botany, University of Lurid, O. Vallgatan 18-20, S-223 61 Lund, Sweden

lations (e.g. Loveless and Hamrick 1984; Epperson 1989), and the spatial hierarchy of genetic diversity is a complex reflection of both selective and stochastic processes. Levels of gene flow will determine the relative contributions of selection and genetic drift to the process of population differentiation (e.g. Levin 1984; Slatkin 1987; Oshawa et al. 1992). Allozyme studies of the structure of genetic diversity reveal that long-lived, outcrossing plants show relatively little differentiation between populations (Hamrick and Godt 1989; Hamrick 1990). Hence, indirect estimates of gene flow rates from such species are usually high (Govindaraju 1988; Ellstrand 1992). Extensive gene flow may limit evolution by hindering adaptation to local conditions (Slatkin 1987). Yet, despite the body of evidence suggesting that gene flow between populations of outcrossing species is extensive, a number of studies have revealed local differentiation within populations of perennial outcrossing plants (e.g. Schaal 1975; Brown 1979; Prentice 1984; Schnabel and Hamrick 1990; Gehring and Linhart 1992). Fine-scale genetic differentiation within populations may be explained by patterns of vegetative spread in clonal species (e.g. Schnabel and Hamrick 1990). However, in non-clonal species, local genetic structure may reflect restricted neighbourhood size (i.e. limited gene flow by seed or pollen) (Schaal 1975, 1980; Bos et al. 1986) and/or local niche-differentiation (Bradshaw 1972; Snaydon and Davies 1976, 1982; Gehring and Linhart 1992; Prentice et al. 1995). If local gene flow is restricted, a patchy genetic structure may arise as a result of drift in small genetic neighbourhoods, even in large populations (Turner et al. 1982). Restricted gene flow may also facilitate selective differentiation between local habitats (Levin 1988). However, if gene flow is extensive, strong differential selection and pronounced fitness differences between genotypes will be necessary to maintain a patchy local genetic structure (Davies and Snaydon 1976; Levin 1984).

OECOLOGIA

Studies of the diploid perennialherb, Gypsophila fastigiata L. on the Baltic island of Oland have shown that little of the total diversity in allozymes is explained by differentiation between sites (tens of kilometres apart) (Prentice and White 1988; Prentice 1992). A low level of geographic differentiation between sites is consistent with the fact that the species is insect pollinated and primarily outcrossing (cf. Hamrick and Godt 1989; Govindaraju 1988). There is also evidence that suggests that variation at the Pgi-2 locus in G.fastigiata may be associated with variation in plant community composition on a scale of tens of meters (Prentice and Cramer 1990). The present study examines the relationship between genetic differentiation and habitat variables within local populations of G. fastigiata and explores whether allozyme-habitat associations may be accompanied by differences in individual reproductive success.

106 (1996) 16o30 '

N

9

57°00 '

OF

KL GY 56°30 '

PE ME AB

Gypsophilajastigiata L. is a long-lived diploid perennial herb which reaches its northern range limit in a number of disjunct regional populations in Fennoscandia (Jalas and Suominen 1986; Prentice 1992) and is widespread in the steppe-like "alvar" grasslands in the southern part of the Baltic island of 0land (Sterner 1938; Bengtsson et al. 1988). Individuals of G. fastigiata form compact cushions with a central tap root and do not spread vegetatively. The species is self-compatible but experiments show that the selfing rate in field populations is less than 1% (H. C. Prentice and M. L6nn, unpublished work).

309

17o00 '

Materials and methods The species

© Springer-Verlag

0

I

10

20kin

I

Fig. 1 The six sampling sites for Gypsophilafastigiata on the Baltic island of 01and (OF Odens flisor (Karums alvar), KL Kleva alvar, GY Gynge alvar, PE Pengtsa, ME Mellby, AB Albrunna)

Habitat, sites and data collection The range of alvar habitats occupied by G. jastigiata is described by Bengtsson et al. (1988). The major gradients in plant community composition in the dry alvar grasslands are associated with variation in soil type and moisture and are also characterized by differences in the total vegetation cover and the separate covers of bryophytes and lichens (Rosdn 1982; Bengtsson et al. 1988). The present study was carried out during 1991 at six sites (Fig. 1), chosen to cover the entire range of alvar habitats occupied by G. fastigiata on Oland. The mean distance between adjacent sites was 10.9 km (Fig. 1). Within sites, c. ten 0.25 m 2 plots were chosen so as to contain at least one adult G. fastigiata individual and to provide a representative sample from the different microhabitats within the sites. Data were collected from a total of 51 plots 9, 7, 10, 6, 11 and 8 plots respectively for sites OF to AB. The mean pairwise distances between plots within sites were: 69 m (OF), 21.5 m (KL), 17.5 m (GY), 13 m (PE), 69.5 m (ME) and 23.5 m (AB). We recorded seven habitat variables in each of the plots. Overall abiotic and biotic habitat variation was characterized by the total percentage cover of vegetation, and the separate percentage covers of vascular plants, grasses and sedges, bryophytes and lichens. Species richness was characterized by the total number of vascular plant species and we measured soil depth at five positions in each plot to obtain mean soil depth. The means and standard errors of the habitat variables are as follows: total vegetation cover 69.3 (13.5), vascular plant cover 58.8 (15.1), grass cover 17.2 (13.9),

bryophyte cover 8.32 (9.67), lichen cover 5.93 (4.81), species richness 11.8 (3.03), soil depth 4.66 (1.62). Data on fruit production were collected from all 241 adult individuals within the 51 plots in mid-August. We used the number of infructescences per individual as a measure of reproductive success. G. fastigiata produces large numbers of seeds (mean 1910, range 35-21408) per plant and an average of 58 (range 7 173) seeds per infructescence (K. Bengtsson, unpublished work). The number of infructescences per individual is positively correlated with seed weight and seed weight is positively associated with germination success and seedling viability (K. Bengtsson and L. P. Lefkovitch, unpublished work). The mean number of infructescences per adult individual in the present study was 8.10 (SE = 18.4).

Enzyme electrophoresis Within each plot, leaf samples were taken from all the G. Jhstigiata individuals with more than four leaves. A total of 387 individuals was screened for electrophoretic variation. Electrophoretic procedures followed L6nn and Prentice (1990), using buffer system i. Staining for phosphoglucose isomerase (PGI, EC 5.3.1.9) and triosephosphate isomerase (TPI, EC 1.6.4.3) was carried out according to Soltis et al. (1983). Staining for superoxide dismatase (SOD,EC 1.15.1.1) was carried out using the TPI recipe, but at pH 8.6 and excluding dihydroxyacetone phosphate.

310

OECOLOGIA

106 (1996)

© Springer-Verlag

All enzymes migrated anodally; loci and alleles were numbered in order of decreasing mobility. Variation was scored at three polymorphic loci: Pgi-2, Sod-1 and Tpi-1. Three alleles [1, 2 and 3 (Prentice 1992)] were present at (dimeric) Pgi-2 in the studied populations. The Tpi-1 locus coded for a dimeric enzyme and had five alleles (lf, 1, 2, 3 and 4), with homodimeric Rfvalues of 0.79, 0.69, 0.64, 0.58 and 0.54 respectively. Variation in the Tpi-2 banding region was not interpreted. The Sod-1 locus coded for a dimeric enzyme with three alleles, 1 (Rf = 0.70), 2 (Rf= 0.66) and 3 (Rf = 0.58). Statistical analyses for Pgi-2 and Tpi-1 were based on all 387 screened individuals, while the analyses for Sod-1 were based on data from 381 individuals.

The structure of allelic diversity Allelic diversities were estimated for each locus using the diversity statistic (H) of Nei (1973) with correction for small sample size according to Nei (1978). The total allelic diversity (/arT) was estimated from the total data set. The mean allelic diversity within sites (Hs) was estimated as the weighted mean of the within-site diversities and the mean allelic diversity within plots (Hp) was estimated as the weighted mean of the within-plot diversities. Allelic diversity was partitioned into its within- and between-site and -plot components using the approach described in Chakraborty et al. (1982). The proportion of the total alMic variation (Hx) due to variation between sites is given by GsT [= (HT -- Hs)/HT], the proportion of variation due to variation between plots within sites is given by Gps [= (Hs - Hp)/Hx] and the proportion of variation due to variation within plots is given by Hp/HT. A test of whether the Gsx and Gps values for the individual loci were larger than those expected by chance, given the sampling structure in the present study, was carried out using a Monte Carlo permutation procedure in which individual genotypes were reassigned randomly to sites and plots. In a first run, testing the between-site differentiation (GsT), all individual genotypes were randomly assigned to sites, keeping the number of individuals per site the same as the original numbers. In a second run, testing between-plot differentiation within sites (Gps), all individual genotypes within each site were randomly assigned to the plots within the site, keeping the number of individuals per plot the same as the original numbers. This approach is similar to that used by Giles and Goudet (1996) for F statistics. Values for GsT and GPs obtained in each of the Monte Carlo permutations were compared with the original values for GsT and Gps. Significance probabilities for GsT and GPs were calculated as the number of times in which the randomised values were equal to or greater than the original values, divided by the number of permutations (= 1000). Tests for deviations from Hardy-Weinberg genotype frequency expectations were carried out on plots and sites for each locus. Where necessary, rare alleles were pooled successively to give "expected" genotype cell values of not less than one. HardyWeinberg statistics were computed using the program M U L E written by R. J. White.

Analysis of associations between allelic variation, fruit production and environmental variables We used generalized linear models to carry out analyses of deviance (McCullagh and Nelder 1989) on the relationships between allelic variation, habitat variables and fruit production. The generalized linear model concept includes the possibility of specifying appropriate error distributions for the individual response variables, rather than transforming the response variables in an attempt to achieve a normal distribution. Response variables based on allele frequencies within individuals (counts with an upper limit) were assumed to have a binomial error distribution and the.response

variable "infructescence number" (counts) was assumed to have a Poisson error distribution (cf. McCullagh and Nelder 1989; Crawley 1993). We used the canonical link functions "logit" for models with a binomial error distribution and "log" for models with a Poisson error distribution. The interpretation of an analysis of deviance follows similar principles to those used in an analysis of variance. The deviance is a measure of the goodness-of-fit of the model to the data (Crawley 1993). The variation explained by a term can be measured by the deviance change caused by the inclusion of the term into a model. The deviance change is usually well approximated by a Z2 distribution (McCullagh and Nelder 1989). We use the )~2 approximation for significance testing in the cases when the dispersion factor (estimated from the mean deviance for the error term) is close to unity, as is expected when the response variables are counts with binomial or Poisson error distributions. In cases of over-dispersion we instead use the mean deviance change divided by the dispersion factor, referring this mean deviance ratio to an F-distribution (Crawley 1993). Our models are constructed so that the explanatory variables are included in the model one after one. The change in deviance is recorded after the inclusion of each new variable. The effects of explanatory variables fitted earlier in the model are thus removed from the data set and do not influence the effects of subsequently-introduced terms. If explanatory variables are correlated with each other, the first-included variable may explain variation that could as well have been explained by the second variable. However, the last-included variable can have an effect of its own, even if it is correlated with a variable that was included earlier in the model. The response variables in the analyses of deviance were either infructescence number (from 241 adult individuals) or electrophoretic data from 387 individuals (381 for Sod-l). Allelic variation was handled as allele counts (0, 1 or 2) per individual. Alleles if and 4 at Tpi-1 each occurred in less than 1% of the individuals and were not included in the analyses based on allele counts. In each analysis of deviance, we included the explanatory variable "site" and the interactions including '"site" as early as possible in the model. After the removal of site effects, the genetic and habitat variables of interest, and their interactions, were then successively included in the analytical model. Because of the possibility that within-plot variation may be spatially autocorrelated as a result of limited seed dispersal (cf. Schnabel and Hamrick 1990), the plants within a plot cannot be regarded as independent observations. For this reason, we treated within-plot variation as subsampling in the analyses and the dispersion factor (estimate of random variation in the model) is estimated from the term "between plots within sites" (dispersion factor = "plot" deviance/df). The allele-habitat analyses focussed on the relationships between the individual alleles, and site and habitat variables. A separate analysis of deviance was carried out for each of the nine alleles as the response variable. The relationships between fruit production, habitat and enzyme variation (for chosen alleles) were investigated in an analysis of deviance with infructescence number as the response variable. As above, the error term is taken from the term "plot". Interaction terms between variates (continuous variables), were calculated as the product of the original variates (Lawes Agricultural Trust 1987). To facilitate the interpretation of the interaction terms, the original variates were centred around zero by subtracting their means before multiplying them (L. Lefkovitch, personal communication). The interaction variate takes a large value when the original variates parallel each other (i.e. are positively correlated) and a low value when they are inversely related. The generalized linear model analyses were carried out using G E N S T A T 5, release 2 (Lawes Agricultural Trust 1987).

OECOLOGIA Table 1 Relative allele frequencies for GypsophiI..a fastigiata at six sites on Oland in 1991 (n = number of individuals, sites shown in Fig. 1)

106 (1996)

© Springer-Verlag

311

Site Locus

Pgi-2

Allele

OF

KL

GY

PE

ME

AB

1

0.000 0.896 0.104 24

0.068 0.780 0.152 59

0.041 0.648 0.311 135

0.043 0.783 0.174 23

0.069 0.771 0.160 109

0.068 0.770 0.162 37

2 3 n

0.000 1.000 0.000 24

0.085 0.890 0.025 59

0.102 0.788 0.110 132

0.000 0.696 0.304 23

0.005 0.841 0.154 107

0.014 0.847 0.139 36

if 1 2 3 4 n

0.021 0.146 0.750 0.083 0.000 24

0.000 0.102 0.788 0.110 0.000 59

0.007 0.089 0.881 0.022 0.000 135

0.000 0.022 0.870 0.109 0.000 23

0.000 0.037 0.894 0.064 0.004 109

0.000 0.094 0.892 0.014 0.000 37

2 3 n

Sod-1

Tpi-~

1

Results

Table 2 H statistics and the partitioning of allelic diversity at the loci Pgi-2, Sod-1 and Tpi-1. Significance tests for departure of Gsv and Gps from random expectations are described in the text

The structure of allelic diversity The allele frequencies at the six sites are shown in Table 1. Single-locus allelic diversities (Hr) are shown in Table 2, along with the mean within site (Hs) and mean within-plot (Hp) diversities. Table 2 also shows the partitioning of allelic diversity at each of the three loci. Although the values for the total genetic diversity vary between the loci, the proportions of the total diversity due to variation within and between plots and sites are similar for the three loci. All values for GsT and Gps are significantly greater than would be expected if the genotypes were randomly distributed between plots and sites (Monte Carlo test, P < 0.01). The majority of the diversity (89-91%) is accounted for by variation within plots. The proportion of the diversity due to differentiation between plots within sites (6-9 %) is 2-7 times greater than that between sites

Locus

Total diversity (Hr) Mean within site diversity ()ffs) Mean within plot diversity (~rp)

Pgi-2

Sod-1

Tpi-1

0.410 0.401 0.364

0.292 0.283 0.265

0.247 0.244 0.223

0.032*** 0.061"** 0.907

0.011"* 0.082*** 0.907

Partitioning of the total diversity between sites (GsT) 0.022*** between plots within sites (Gps) 0.089*** within plots (Hp/Hv) 0.889 ** P < 0.01, *** P < 0.001

tations (expected number of deviations caused by "type 1 error" with 95% probability = 7.65 for 3 loci and 51 plots).

(1-3%). Only one site (KL) showed a significant (P < 0.01) deviation of genotype frequencies from HardyWeinberg expectations (at the Tpi-1 locus, with an excess of 2/2 homozygotes and fewer than expected 1/2 heterozygotes), Only three plots showed withinplot deviations (P < 0.05) from Hardy-Weinberg expec-

Alleles, sites and habitat The correlation coefficients between the habitat variables are shown in Table 3. The nine analyses of deviance in Table 4 show the associations between

Table 3 Pearson product-moment correlation coefficients between the environmental variables used to describe habitat variation of

G. fastigiata.

Total vegetation cover (Totveg) Vascular plant cover (Vasc) Grass cover (Grass) Bryophyte cover (Bryo) Lichen cover (Lich) Soil depth (Soil) *P < 0.05, **P < 0.01, ***P < 0.001

Species richness

Totveg

Vasc

Grass

Bryo

Lich

0.36** 0.41"* 0.18 --0.02 0.21 0.03

0.81"** 0.39** --0.04 0.42** --0.13

0.45*** 0.14 0.31' 0.13

0.10 0.47*** 0.32*

--0.02 0.53***

-0.12

312

OECOLOGIA

Table 4 Associations between alleles, sites and habitat characteristics in G. fastigiata: separate analyses of deviance for nine alleles at three loci. The response variables in the analyses are allele counts in individuals, with an assumed binomial distribution. The explanatory variables (site and seven descriptors of habitat) are given in their order of inclusion in the analyses. The dispersion factor is estimated from the mean deviance between plots within sites (i.e. the plot term) and the deviances are divided by the dispersion factor before testing for significance by referring them to a )¢2 distribution. The signs of the regression coefficients for the significant variates are given in parentheses

106 (1996)

© Springer-Verlag

Source of variation

df (change)

Site Species richness Total vegetation cover Vascular plant cover Grass cover Bryophyte cover Lichen cover Soil depth Plot Residual Dispersion factor

5 1 1 1 1 1 1 1

df (change)

Plot Residual Dispersion factor

5 1 1 1 1 1 1 1 45 323

(change)

Plot Residual Dispersion factor

Pgi-2-3

Deviance change

Deviance change

Deviance change

5 1 1 1 1 l 1 1

26.307* 0.698 5.889 0.090 6.964 1.153 22.906*** (+) 0.000

92.758 328.760 2.061

84.827 312.627 1.885

Sod-l-1

Sod-1-2

Sod-l-3

Deviance change

Deviance change

Deviance change

31.840*** 0.010 1.620 3.828 0.621 2.590 0.523 7.384* (+)

39.703*** 0.511 2.970 0.948 2.983 0.000 0.654 12.276"* ( - )

58.894 88.514 1.309

67.165 305.663 1.493

83.213 195.604 1.849

Tpi-l-1

Tpi-l-2

Tpi-l-3

Deviance change

Deviance change

Deviance change

13.118 10,906" (+) 5.613 1.175 4.001 0.031 6.143 1.122

45 329

21.268 4.474 3.066 0.213 10.054' (+) 1.041 27.053*** ( - ) 0.041

56.639 132.752 1.259

41.603*** 0.781 5.493* ( - ) 1.685 0.003 2.438 0.058 1.466

df Site Species richness Total vegetation cover Vascular plant cover Grass cover Bryophyte cover Lichen cover Soil depth

Pg#2-2

7.916 7.265* (+) 1.249 0.069 3.495 0.055 2.786 0.349

45 329

Site Species richness Total vegetation cover Vascular plant cover Grass cover Bryophyte cover Lichen cover Soil depth

Pgi-2-1

80.617 147.446 1.792

13.572 9.398* (+) 4.679 0.518 6.161 0.325 1.601 2.628 87.955 261.149 1.954

19.018 0.296 0.195 0.080 0.870 0.167 0.328 8.871 * ( - ) 78.806 109.830 1.751

*P < 0.05, **P < 0.01, ***P < 0.001

allele counts (for the individual alMes) and site and habitat variables. Allele counts differ significantly between sites for Pgi-2-3 and for the three Sod-1 alleles. Several alleles show significant associations with habitat variables, after the removal of site effects. At the Pgi-2 locus, allele 1 is positively associated with species richness, allele 2 is positively associated with grass cover and negatively associated with lichen cover, while allele 3 is positively associated with lichen cover. At Sod-i, allele 1 is negatively associated with total vegetation cover. Sod-l-2 is positively associated with,

and Sod-l-3 negatively associated with soil depth. Tpi-1 alleles 1 and 2 are positively associated with species richness and allele 3 is negatively associated with soil depth. Grass cover is correlated with lichen cover and soil depth (Table 3), so it is possible that the effect of grass cover may be at least partly explained by the correlated terms included later in the model. The same is true for species richness and total vegetation cover which are also correlated with some of the variables included later in the models (Table 3).

OECOLOGIA Table 5 Associations between fruit production, sites, habitat characteristics and alleles in G. fastigiata. The analysis of deviance had infiuctescence number in adult individuals (with an assumed Poisson distribution) as the response variable, and included site, the seven descriptors of habitat and allele counts (per individual) for the alleles Pgi-2-2, Pgi-2-3 and Sod-l-3 as explanatory variables. The explanatory variables and their interactions are given in their order of inclusion in the model. Details of the handling of interactions between variates are given in the text. The dispersion factor is estimated fi-om the deviance between plots within sites (i.e. the plot term). Mean deviance ratios (mean deviance/dispersion factor) are tested for significance by referring them to an F-distribution. The regression coefficients for the significant variates are given in parentheses Source of variation Site (S) Species richness Total vegetation cover Vascular plant cover Grass cover Bryophyte cover Lichen cover Soil depth S × Species richness S × Total vegetation cover S x Vascular plant cover S × Grass cover S x Bryophyte cover S x Lichen cover S x Soil depth

Pgi-2-2 Pgi-2-3 Sod-1-3 S × Pgi-2-2 S x Pgi-2-3 S x Sod-1-3 Pgi-2-2 × Lichen cover Pgi-2-3 × Lichen cover Sod-1-3 x Soil depth S x Pgi-2-2 x Lichen cover S × Pgi-2-3 × Lichen cover S x Sod-1-3 x Soil depth Plot Residual Dispersion factor

df

Deviance (change) change 5 1 1 1 1 1 1 1 5 5 5 5 4 4 3 1 1 1 5 4 4 1 1 1 5 3 3

625.58 6.01 1.05 7.46 9.01 44.24 42.18 5.17 661.16 237.18 227.21 153.12 91.29 33.36 55.59 27.36 0.23 0.21 242.64 71.70 58.8i 98.41 78.08 27.83 11.17 36.78 64.83

21 146

220.77 1467.04 10.51

Mean deviance ratio 11.90"** 0.57 0.10 0.71 0.86 4.21 4.01 0.49 12.58"** 4.51"* 4.32** 2.91 2.17 0.79 1.76 2.60 0.02 0.02 4.62** 1.71 1.40 9.36** ( - 5 . 5 ) 7.43* ( - 1.0) 2.65 0.21 1.17 2.06

*P < 0.05, **P < 0.01, ***P < 0.001

Soil depth is included last in the models, so the effect of soil depth is an effect of this term after the removal of the effects of all other habitat variables. Lichen cover is included as the last-but-one term. Because lichen cover is not correlated with soil depth (Table 3), the effect of lichen cover is thus an effect of this term by itself. Fruit production, habitat and alleles Table 5 shows an analysis of deviance which examined the associations between infi'uctescence number, site

106 (1996)

© Springer-Verlag

313

and habitat characteristics. In addition, the model included those alleles that showed significant (P < 0.01) associations with habitat variables in Table 4 and interactions between the alleles and the habitat variables with which they were significantly associated in Table 4. Limitations of program capacity and degrees of freedom (and the problem of allele collinearity) prevented the inclusion of all nine alleles and their interactions with site membership and the seven habitat variables into the model. Fruit production was significantly associated with site (cf. Table 1) and with the site x habitat interactions for species richness, total vegetation cover and vascular plant cover. These interactions indicate that infructescence number is related to habitat variation, but that the nature of the associations with the different habitat descriptors differs between sites. Fruit production also showed a significant site × Pgi-2-2 interaction, indicating that the relationship between fruit production and allele 2 at Pgi-2 differs between sites. There were also significant Pgi-2-2 and Pgi-23 x lichen cover interactions. Overall fruit production is low when counts of alleles 2 or 3 and lichen cover are positively correlated, and fruit production is high when counts of alleles 2 or 3 and lichen cover are negatively correlated. The negative regression coefficient was 5 times larger for allele 2 than for allele 3.

Discussion The spatial structure of allozyme variation The hierarchical partitioning of genetic diversity in the present study shows that a low, but significant proportion (c. 2%, see Table 2) of the total allelic diversity in G. fastigiata is accounted for by variation between sites (up to 50 km apart). Relatively low levels of differentiation between populations are characteristic of many outcrossing perennial species (Hamrick and Godt 1989; Hamrick et al. 1991), and low levels of population differentiation have been taken as indirect evidence for extensive gene flow (e.g. Govindaraju 1988; Ellstrand 1992). The sampling structure in the present study, in addition to the site level, included a lower spatial level that of plots, separated by 10-70 m, within sites. Despite the fact that only one locus at one site showed genotype deviations from Hardy-Weinberg expectations, the proportion of the total diversity due to variation between plots within sites is approximately 3 times greater than the proportion due to variation between sites (Table 2). The between-plot variation indicates a degree of spatial patchiness within sites. However, the low levels of between-site allozyme differentiation - on a larger spatial scale - suggest that the within-site differentiation is unlikely to be a simple reflection of restricted gent flow.

314

O E C O L O G I A 106 (1996) © Springer-Verlag

Allelic variation and habitat variation The alvar grasslands occupied by G.fastigiata on 01and are characterized by a high diversity of both species and plant communities (Bengtsson et al. 1988; van der Maarel 1988; van der Maarel and Sykes 1993). Within sites, there is a complex mosaic of habitat types, with spatial patchiness on scales of decimetres to tens of metres (cf. Bell et al. 1991). The range of habitats occupied by G. fastigiata is similar in each of the six studied sites. Our results show that most alleles are significantly associated with within-site variation in vegetation cover or soil depth (Table 4). The strongest relationship is that between lichen cover and variation at the Pgi-2 locus. In the grassland habitats occupied by G. fastigiata (cf. Fig. 4 in Bengtsson et al. 1988), high lichen cover is associated with low soil moisture. Allele 3 at Pgi-2 can thus be tentatively linked to drier situations and allele 2 to moister habitats. Several studies have shown relationships between habitat variation and allozyme variation in plants (e.g. Hamrick and Allard 1972; Lumaret 1984; Mitton et al. 1989; L6nn 1993) and a previous study of G. fastigiata on {)land showed an association between variation at Pgi-2 and gradients of variation in plant community composition in the alvar grasslands (Prentice and Cramer 1990). Data from both plant and animal studies suggest that selection may act directly on allozyme genotypes (e.g. Brown et al. 1976; Zangerl and Bazzaz 1984; Shihab and Heath 1987; Powers et al. 1991; Riddoch 1993). Associations between allozyme variation and habitat may also reflect indirect selection and hitchhiking (Hedrick 1980; Vrijenhoek et al. 1992). However, in outcrossing species such as G. fastigiata, only strongly-interacting loci are likely to show substantial levels of linkage disequilibrium (Hastings 1989).

Interactions between alleles, habitat and reproductive SUCCESS

If the allele-habitat associations in Table4 reflect differential selection on adult individuals in different local habitats, it is predicted that individuals with the "right" allele-habitat combination should be relatively more fit than the individuals with the "wrong" allelehabitat combination. This prediction was tested by using fruit production as a measure of individual fitness, and using the three strongest allele-habitat associations (in Table 4) as explanatory variables. After the removal of the direct effects of site, habitat and alleles, two of the three allele x habitat combinations (both at Pgi-2) showed a significant negative association with fruit production (Table 5). In other words, when an allele occurs in an environment where it is rare - in relation to other environments - it occurs

in less fit individuals. Allele 2 is relatively rare in lichenrich habitats and is associated with lower fruit production in lichen-rich plots. Individual fitness in G. fastigiata is apparently influenced by allele-environment interactions at Pgi-2. Associations between variation at PGI and fitness-related characters have previously been shown in studies on a range of different organisms (e.g. Shihab and Heath 1987; Jacobs 1990; Wolff and Haeck 1990; Bush and Smouse 1991; Watt 1992). Genotype x environment interactions influence relative fitness at different stages in the plant life cycle (Ennos 1989; Stratton 1992). Differential fruit production may reflect the vigour of different genotypes in different habitats and provide a measure of the performance of particular allele-habitat combinations. In G. fastigiata infructescence number is positively correlated with seed size. Seed size is, in its turn, associated with germination rate, germination success and seedling viability (K. Bengtsson and L. Lefkovitch, unpublished work). G. fastigiata is a long-lived perennial with a mean half-life of 5-75 years for adult individuals (K. Bengtsson, unpublished work). Seedling establishment occurs in gaps within a matrix of already established individuals. Seed dispersal mostly occurs within a radius of less than 20 cm from maternal individuals (K. Bengtsson, unpublished work). Restricted seed dispersal is expected to skew the local seed pool in favour of maternal alleles and, together with differential fruit production in different allele x habitat combinations, serve to strengthen the pattern of allele-habitat associations within sites. The low level of differentiation between {)land populations of G. fastigiata is interpreted as a reflection of the homogenizing effect of extensive gene flow by pollen, presumably over many generations. In contrast, field observations suggest that gene flow by seed may be extremely restricted in G. fastigiata. The effect of gene flow by pollen may be locally overridden by limited seed dispersal (Schnabel and Hamrick 1990) or if individuals with favourable allele x habitat combinations have a higher reproductive fitness. Acknowledgements We would like to thank Leonard Lefkovitch for introducing us to generalized linear models, and Stefan Andersson, Norman Ellstrand, Tomas Pfirt and two anonymous reviewers for commenting on the manuscript. The study was funded by grants (to all three authors) from the Uppsala University Ecological Research Station, Olands Skogsby, and (to H.C.P.) from the Swedish research councils NFR and SJFR and the Swedish Environmental Protection Agency (SNV).

References Bell G, Lechowicz MJ, Schoen DJ (1991) The ecology and genetics of fitness in forest plants. III. Environmental variance in natural populations of Impatiens pallida. J Ecol 79:697-713 Bengtsson K, Prentice HC, Ros6n E, Moberg R, Sj6gren E (1988) The dry alvar grasslands of Oland: ecological amplitudes of plant

O E C O L O G I A 106 (1996) species in relation to vegetation composition. Acta Phytogeogr Suec 76:21-46 Bos M, Harmens H, Vrieling K (1986) Gene flow in Plantago. I. Gene flow and neighbourhood size in P. lanceolata. Heredity 56:43 54 Bradshaw AD (1972) Some of the evolutionary consequences of being a plant. Evol Biol 5:25-47 Brown AHD (1979) Enzyme polymorphism in plant populations. Theor Popul Biol 15:1-42 Brown AHD, Marshall DR, Munday J (1976) Adaptedness of variants at an alcohol dehydrogenase locus in Bromus mollis L. (soft bromegrass). Aust J Biol Sci 29:389-396 Bush RM, Smouse PE (1991) The impact of electrophoretic genotype on life history traits in Pinus taeda. Evolution 45:481-498 Chakraborty R, Haag M, Ryman N, Stfihl G (1982) Hierarchical gene diversity analysis and its applications to brown trout population data. Hereditas 97:17 21 Crawley MJ (1993) GLIM for ecologists. Blackwell Scientific Publications, Oxford Davies MS, Snaydon RW (1976) Rapid population differentiation in a mosaic environment. III. Measures of selection pressures. Heredity 36:59-66 Ellstrand NC (1992) Gene flow by pollen: implications for plant conservation genetics. Oikos 63:77-86 Ennos RA (1989) Detection and measurement of selection: genetic and ecological approaches. In: Brown AHD, Clegg MT, Kahler AL, Weir BS (eds) Plant population genetics, breeding, and genetic resources. Sinauer, Sunderland, pp 200-214 Epperson BK (1989) Spatial patterns of genetic variation within plant populations. In: Brown AHD, Clegg MT, Kahler AL, Weir BS (eds) Plant population genetics, breeding, and genetic resources. Sinauer, Sunderland, pp 229-253 Gehring JL, Linhart YB (1992) Population structure and genetic differentiation in native and introduced populations of Deschampsia caespitosa (Poaceae) in the Colorado Alpine. Am J Bot 79:1337-1343 Giles BE, Goudet J (1996) Genetic differentiation in Silene dioica metapopulations: estimation of spatiotemporal effects in a successional plant species, ban Nat (in press) Govindaraju DR (1988) Relationship between dispersal ability and levels of gene flow in plants. Oikos 52:31-35 Hamrick JL (1990) Isozymes and the analysis of genetic structure in plant populations. In Soltis DE, Soltis PS (eds) Isozymes in plant biology. Chapman and Hall, London, pp 87-105 Hamrick JL, Allard RW (1972) Microgeographical variation in allozyme frequencies in Arena barbata. Proc Natl Acad Sci USA 69:2100-2104 Hamrick JL, Godt MJW (1989) Allozyme diversity in plant species. In: Brown AHD, Clegg MT, Kahler AL, Weir BS (eds) Plant population genetics, breeding, and genetic resources. Sinauer, Sunderland, pp 43-63 Hamrick JL, Godt MJW, Murawski DA, Loveless MD (1991) Correlations between species traits and allozyme diversity: implications for conservation biology. In: Falk DA, Holsinger KE (eds) Genetics and conservation of rare plants. Oxford University Press, Oxford, pp 75-88 Hastings A (1989) The interaction between selection and linkage in plant populations. In: Brown AHD, Clegg MT, Kahler AL, Weir BS (eds) Plant population genetics, breeding, and genetic resources. Sinauer, Sunderland, pp 163-180 Hedrick PW (1980) Hitchhiking: a comparison of Iinkage and partial selfing. Genetics 94:791-808 Jacobs J (1990) Microevolution in predominantly clonal populations of pelagic Daphnia (Crustacea: Phyllopoda): selection, exchange, and sex. J Evol Biol 3:257-282 Jalas J, Suominen J (1986) Atlas Florae Europaeae, vol. 7. The Committee for Mapping the Flora of Europe/Societas Biologica Fennica Vanamo, Helsinki, Finland Lawes Agricultural Trust (1987) Genstat 5 manual. Genstat 5 Committee, Rothamstead/Clarendon Press, Oxford

© Springer-Verlag

315

Levin DA (1984) Immigration in plants: an exercise in the subjunctive. In Dirzo R, Sarukhfin J (eds), Perspectives on plant population ecology. Sinauer, Sunderland, pp 242-260 Levin DA (1988) Local differentiation and the breeding structure of plant populations. In: Gottlieb LD, Jain SK (eds), Plant evolutionary biology. Chapman and Hall, London, pp 305 329 L6nn M (1993) Genetic structure and allozyme-microhabitat associations in Bromus hordeaceus L. Oikos 68:99-106 L6nn M, Prentice HC (1990) Mosaic variation in Swedish Petrorhagia prolifera (Caryophyllaceae): the partitioning of morphometric and electrophoretic diversity. Biol J Linn Soc 41: 353 373 Loveless MD, Hamrick JL (1984) Ecological determinants of genetic structure in plant populations. Annu Rev Ecol Syst 15: 65-95 Lumaret R (1984) The role of polyploidy in the adaptive significance of polymorphism at the GOT 1 locus in the Daclylis glomerata complex. Heredity 52:153 169 Maarel E van der (1988) Floristic diversity and guild structure in the grasslands of C)lands Stora Alvar. Acta Phytogeogr Suec 76: 53-65 Maarel E van der, Sykes MT (1993) Small-scale plant species turnover in grasslands: the carousel model and some comments on the niche concept. J Veg Sci 4:179-88 McCullagh P, Nelder JA (1989) Generalized linear models. Chapman and Hall, London Mitton JB, Stutz HP, Schuster WS, Shea KL (1989) Genotypic differentiation at PGM in Engelmann spruce from wet and dry sites. Silvae Genet 8:217-222 Nei M (1973) Analysis of gene diversity in subdivided populations. Proc Natl Acad Sci USA 70:3321-3323 Nei M (1978) Estimation of average heterozygosity and genetic distance fi'om a small number of individuals. Genetics 89: 583 590 Oshawa R, Furuya N, Ukai Y (1992) Effect of spatially restricted pollen flow on spatial genetic structure of an animal pollinated allogamous plant population. Heredity 71:64 73 Powers DA, Lauerman T, Crawford D, DiMichele L (1991) Genetic mechanisms for adapting to a changing environment. Annu Rev Ecol Syst 25:629-659 Prentice HC (1984) Enzyme polymorphism, morphometric variation and population structure in a restricted endemic, Silene diclinis (Caryophyllaceae). Biol J Linn Soc 22: 125-143 Prentice HC (1992) The structure of morphometric and allozyme variation in relict populations of GypsophilafiTstigiata in Sweden. Biol J Linn Soc 47:197-216 Prentice HC, Cramer W (1990) The plant community as a niche bioassay: environmental correlates of local variation in Gypsophila fastigiata. J Ecol 78:313-325 Prentice HC, White RJ (1988) Variability, population size and isolation: the structuring of diversity in Oland Gypsophila fast# giata. Acta Oecol Oecol Plant 9:19-29 Prentice HC, L6nn M, Lefkovitch LP, Runyeon H (1995) Associations between allele frequencies in Festuca ovina and habitat variation in the alvar grasslands on the Baltic island of Oland. J Ecol 83:391-402 Riddoch BJ (1993) The adaptive significance of electrophoretic mobility in phosphoglucose isomerase (PGI). Biol J Linn Soc 50:1-17 Ros6n E (1982) Vegetation and sheep grazing in limestone grasslands of south Oland, Sweden. Acta Phytogeogr Suec 72:1-108 Schaal BA (1975) Population structure and local differentiation in Liatris cylindracea. Am Nat 109 : 511-528 Schaal BA (1980) Measurement of gene flow in Lupinus texensis. Nature 284:450-451 Schnabel A, Hamrick JL (1990) Organization of genetic diversity within and between populations of Gleditsia triacanthos (Leguminosae). Am J Bot 77:1060-1069

316

OECOLOGIA

106 (1996)

© Springer-Verlag

Shihab AF, Heath DJ (1987) Components of fitness and the PGI polymorphism in the freshwater isopod Asellus aquaticus (L.) 2. Zygotic selection. Heredity 58:289-295 Slatkin M (1987) Gene flow and the geographic structure of natural populations. Science 236:787-792 Snaydon RW, Davies MS (1976) Rapid population differentiation in a mosaic environment. IV. Populations of Anthoxanthum odoratum at sharp boundaries. Heredity 37:9-25 Snaydon RW, Davies TM (1982) Rapid divergence in plant populations in response to recent changes in soil conditions. Evolution 36:289~97 Soltis DE, Haufler CH, Darrow DC, Gastony GJ (1983) Starch gel electrophoresis of ferns: a compilation of grinding buffers, gel and electrode buffers, and staining schedules. Am Fern J 73:9 27 Sterner R (1938) Flora der Insel Oland. Acta Phytogeogr Suec 9: 5-169 Stratton DA (1992) Life-cycle components of selection in Erigeron annuus: II. Genetic variation. Evolution 46:107-120

Turner ME, Stephens JC, Anderson WW (1982) Homozygosity and patch structure in plant populations as a result of nearestneighbor pollination. Proc Natl Acad Sci USA 79:203-207 Vrijenhoek RC, Pfeiler E, Wetherington JD (1992) Balancing selection in a desert stream-dwelling fish, Poeciliopsis monacha. Evolution 46:1642-1657 Watt WB (1992) Eggs, enzymes, and evolution: natural genetic variants change insect fecundity. Proc Natl Acad Sci USA 89:10608-10612 Wolff K, Haeck J (1990) Genetic analysis of ecologically relevant morphological variability in Plantago lanceolata L. VI The relation between allozyme heterozygosity and some fitness components. J Evol Biol 3:243-255 Zangerl AR, Bazzaz FA (1984) Niche-partitioning between two phosphoglucoisomerase genotypes in Arnaranthus retroflexus. Ecology 65:218-222