The role of genetic structure in the adaptive divergence of populations experiencing saltwater intrusion due to relative sea-level rise

Authors

  • K. M. Purcell,

    Corresponding author
    1. Beaufort Laboratory, NOAA: Southeast Fisheries Science Center, Beaufort, NC, USA
    • Department of Biology, University of Louisiana at Lafayette, Lafayette, LA, USA
    Search for more papers by this author
  • A. Hitch,

    1. Department of Biology, University of Louisiana at Lafayette, Lafayette, LA, USA
    2. Museum of Wildlife and Fish Biology, University of California at Davis, Davis, CA, USA
    Search for more papers by this author
  • S. Martin,

    1. Department of Biology, University of Louisiana at Lafayette, Lafayette, LA, USA
    2. Apalachicola Field Laboratory, Florida Fish and Wildlife Conservation Commission, Fish and Wildlife Research Institute, Eastpoint, FL, USA
    Search for more papers by this author
  • P. L. Klerks,

    1. Department of Biology, University of Louisiana at Lafayette, Lafayette, LA, USA
    Search for more papers by this author
  • P. L. Leberg

    1. Department of Biology, University of Louisiana at Lafayette, Lafayette, LA, USA
    Search for more papers by this author

Correspondence: Kevin M. Purcell, Beaufort Laboratory, NOAA: Southeast Fisheries Science Center, 101 Pivers Island Road, Beaufort, NC 28516-9722, USA. Tel.: +1 252 728 8761; fax: +1 252 728 8619;

e-mail: kevin@kevin-purcell.com

Abstract

Saltwater intrusion into estuaries creates stressful conditions for nektonic species. Previous studies have shown that Gambusia affinis populations with exposure to saline environments develop genetic adaptations for increased survival during salinity stress. Here, we evaluate the genetic structure of G. affinis populations, previously shown to have adaptations for increased salinity tolerance, and determine the impact of selection and gene flow on structure of these populations. We found that gene flow was higher between populations experiencing different salinity regimes within an estuary than between similar marsh types in different estuaries, suggesting the development of saline-tolerant phenotypes due to local adaptation. There was limited evidence of genetic structure along a salinity gradient, and only some of the genetic variation among sites was correlated with salinity. Our results suggest limited structure, combined with selection to saltwater intrusion, results in phenotypic divergence in spite of a lack of physical barriers to gene flow.

Introduction

Current estimates indicate sea-level rise during the twentieth century has been 3.1 mm year−1 (Nicholls et al., 2007). The northern coast of the Gulf of Mexico is experiencing relative sea-level rise at approximately 10 times global averages (Penland & Ramsey, 1990). Relative sea-level rise in coastal Louisiana is dramatically higher than global averages due to subsidence, alterations in freshwater input, reduced sedimentation rate and degradation of marsh habitat. As a result of these habitat changes, saltwater has gained access to previously protected marsh areas, resulting in increased marsh inundation and saltwater intrusion events. These saltwater intrusion events alter the physical and abiotic characteristics of the marsh environment, creating stressful conditions for nektonic species.

The adaptation of populations to environmental changes over ecological timescales is now widely accepted (Stockwell et al., 2003; Kinnison et al., 2007). Local adaptation is a condition in which a population possesses a higher fitness for the local environment relative to populations from other environments (Kawecki & Ebert, 2004). Under this adaptation model, differences in the characteristics of heterogeneous environments act as stressors, creating locally unique selective pressures for populations (Hedrick, 1986). These selective pressures result in divergent natural selection that alters the allele frequencies among local populations. The occurrence of local adaptation of populations is greatly affected by gene flow, genetic drift, environmental heterogeneity and limitations in both genetic variation and genetic architecture (Kawecki & Ebert, 2004). If gene flow between local populations experiencing different selective environments is high, or if local populations are small enough to experience substantial drift, development of local adaptation will be impeded unless selection is strong (Lenormand, 2002; Kawecki & Ebert, 2004; Orr, 2005).

The western mosquitofish, Gambusia affinis, is a small, sexually dimorphic, Poeciliid fish and a dominant resident of the coastal estuaries of the Gulf of Mexico (Pyke, 2005; Hitch et al., 2011). Whereas the western mosquitofish is a freshwater species, studies have shown that it is ubiquitous across the salinity gradient in coastal marshes (Meffe & Snelson, 1989) and constitutes a considerable proportion of the nektonic biomass in Louisiana marshes (Hitch et al., 2011). This species has been the subject of numerous studies evaluating the impact of salinity stress on fitness (Chervinski, 1983; Congdon, 1994; Nordlie & Mirandi, 1996).

We recently reported differences between western mosquitofish populations in response to salinity stress (Purcell et al., 2008) and demonstrated that they had a genetic basis. Populations of G. affinis from marshes with elevated salinity exposure had higher thresholds for salinity stress, and these thresholds were proportional to previous exposure. In Purcell et al., (2008), we hypothesized that salinity tolerance had evolved in spite of high gene flow. However, conventional theory predicts that the development of differentiated phenotypes is a product of opposing forces increasing differentiation through adaptation to local environments and homogenization resulting from gene flow. Previous studies using allozyme markers in both G. affinis and its sibling species, G. holbrooki (Wooten et al., 1988), have detected genetic differentiation between populations separated by distances of under a kilometre (Smith et al., 1983; McClenaghan et al., 1985). Although rapid dispersal into vacant habitats has been reported (Brown, 1987), limited information is available on dispersal or gene flow, especially in coastal wetlands. The restricted nature of gene flow observed in previous studies of Gambusia suggests that mosquitofish have a genetic structure conducive to the development of local adaptation. Congdon (1994) studied variation at the GPI-2 locus and found that salinity-induced mortality maintained allele frequency differences between populations that were undergoing different levels of selection. These findings indicate that Gambusia are capable of developing well-defined genetic structures across spatially heterogeneous landscapes and that these genetic structures may reflect differences in selective forces. However, Stearns & Sage (1980) report that gene flow from a nearby site appeared to be sufficient to prevent adaptation of a population of G. affinis to local salinities.

Is gene flow in coastal marshes sufficiently limited to aid in the development of local adaptation to salinity, or as we hypothesized in Purcell et al., (2008), did salinity tolerance evolve in spite of high gene flow? Our objective was to characterize the genetic structure in coastal populations to better assess whether limited gene flow may have played a role in the development of adaptively diverged populations in response to sea-level rise. Understanding the degree to which the distribution of genetic variation is associated with historical salinity regimes will allow us to examine patterns of genetic and phenotypic divergence between populations and better understand how evolutionary processes are influencing adaptations to changing sea levels.

Materials and methods

To evaluate the relationship between adaptive divergence and genetic structure, we sampled G. affinis individuals at two spatial levels. First, we collected G. affinis individuals from fresh and brackish marsh locations from two separate estuaries, in Cameron Parish and Terrebonne Parish, Louisiana, USA (Fig. 1a; Table 1). Second, we collected individuals from 24 locations along a salinity gradient in a single estuary in south-east Louisiana (Fig. 1b; Table 2). The objectives of our sampling scheme were to survey sites (i) from phenotypically diverged populations in two spatially distinct estuaries along the Louisiana coast; and (ii) from sites within the same estuary, along a naturally occurring salinity gradient. For this second comparison, we chose to focus on our previous study area in south-eastern Louisiana because differences in salinity in the south-western Louisiana site were partially influenced by levees which could reduce fish movement.

Figure 1.

(a) Map of sampled locations in Cameron Parish (East) and Terrebonne Parish (West) utilized for the between-estuaries comparison. The Atchafalaya River is identified due to its potential role as a barrier to gene flow. (b) Map of sampled locations along a salinity gradient in south-east Louisiana (Terrebonne Parish). The marsh types associated with sampled locations are fresh (F-F9), intermediate (I10-I16) and brackish (B17-B24).

Table 1. Summary of genetic characteristics of fish collected from separate basins including sample size (n), average number of alleles (A), allelic richness (AR), both observed (HO) and expected (HHW) heterozygosity and the inbreeding coefficient (FIS). All values are based on averages from 8 microsatellite loci
SampleSpatial coordinatesnAARHHWHOFIS
LatLon
Fresh
SF29.88−93.533012.876.830.7830.8030.006
HF29.52−90.803011.875.830.720.76−0.065
Brackish
SB29.87−93.513013.127.760.800.83−0.026
HB29.28−91.113012.006.570.710.74−0.032
Table 2. Summary of genetic characteristics of fish collected within a basin including sample size (n), average number of alleles (A), allelic richness (AR), both observed (HO) and expected (HHW) heterozygosity and the inbreeding coefficient (FIS). All values are based on averages from 8 microsatellite loci
SampleSpatial coordinatesnAARHHWHOFIS
LatLon
Fresh
F129.43−90.793011.876.760.780.770.005
F229.44−90.883012.255.730.810.800.015
F329.43−90.873010.756.850.730.670.078
F429.44−90.793011.627.020.700.70−0.012
F529.43−90.813011.626.760.700.690.002
F629.44−90.833013.508.020.780.79−0.012
F729.44−90.823012.627.010.720.700.023
F829.40−90.833011.626.810.680.70−0.032
F929.52−90.803011.876.810.730.76−0.039
Intermediate
I1029.34−90.883011.376.780.750.79−0.045
I1129.33−90.893011.126.730.730.73< 0.001
I1229.37−90.873012.007.400.770.87−0.125
I1329.37−91.112910.506.650.720.79−0.089
I1429.36−91.093011.757.130.760.76−0.010
I1529.36−91.032911.376.770.730.75−0.027
I1629.35−91.023011.507.190.760.80−0.053
Brackish
B1729.28−91.113012.006.880.720.74−0.020
B1829.31−91.112410.756.910.720.77−0.064
B1929.30−91.112410.376.580.700.71−0.021
B2029.33−91.103011.626.870.770.83−0.081
B2129.32−91.503012.126.850.730.76−0.042
B2229.29−91.053012.377.020.730.710.031
B2329.28−91.043011.506.600.690.71−0.027
B2429.28−91.053011.506.370.680.640.053

Characterizing a specific site's salinity is complicated because of daily and seasonal fluctuations in the environmental salinity of estuarine environments. To compensate for this variability, as in Purcell et al., (2008), we used a vegetative classification system (Visser et al., 1998, 2000) to delineate three marsh types based on differences in annual environmental salinity (fresh 0‰, intermediate 2–4‰ and brackish 4–8‰). To evaluate genetic diversity between estuaries, we sampled 30 individuals from fresh and brackish marsh types within two estuaries located approximately 330 km apart (Fig. 1a). Site designations are those used in Purcell et al., (2008). To evaluate genetic diversity within an estuary, we sampled 7–9 sites in each marsh type (fresh, intermediate and brackish) in Terrebonne Parish. We sampled 24–30 individuals from each site for a total of 706 individuals (Table 1). All samples were collected using a drop trap deployed from the bow of an airboat. This method allowed us to sample the small ponds and surface streams located in the internal regions of the marsh. Samples were frozen in the field and stored at −80 °C until processing.

Genomic DNA was extracted using the Puregene tissue extraction protocol (Gentra Systems, Minneapolis, MN, USA). We assayed all samples for seven microsatellite loci developed from G. affinis (Spencer et al., 1999) and one locus developed from its congener Gambusia holbrooki (Zane et al., 1999). The PCR conditions followed those detailed by Spencer et al., (1999) with modified annealing temperatures for Gafμ 1, 5: 54 °C; for Gafμ 2, 3, 4, 6, 7 and Mf-13: 59 °C. Amplification products were analysed in multiplexes of two loci on an ABI 3130 (Applied Biosystems, Carlsbad, CA, USA). Electropherograms were scored using the program genemapper 3.7 (Applied Biosystems). All genotypes were reviewed to ensure accuracy. Both data sets were evaluated for deviations from Hardy–Weinberg equilibrium (HWE) and the presence of linkage disequilibrium (LD) using genepop 3.4 (Raymond & Rousset, 1995). micro-checker 2.2.3 (Oosterhout et al., 2004) was used to test for the presence of null alleles.

Genetic structure within and between basins

In a previous study (Purcell et al., 2008), we examined phenotypic differences in salinity tolerance among G. affinis populations from fresh and brackish marshes in two estuaries in southern Louisiana. Our experimental design assumed that geographical distance together with inhospitable habitat, provided by the Atchafalaya River (Fig. 1a), functioned as a barrier to migration and gene flow, which could create broad-scale genetic differentiation and possibly independent development of local adaptations for salinity tolerance. We tested the validity of that assumption by quantifying the genetic differentiation between estuaries using a microsatellite data set. To evaluate variation in the genetic diversity among and between sampled locations, we calculated pairwise FST (Weir & Cockerham, 1984) estimates using the program fstat 2.9.3 (Goudet, 1995).

Genetic structuring along a salinity gradient

To evaluate the genetic structure of populations within the same estuary, we used the program fstat 2.9.3 (Goudet, 1995) to determine three measures of genetic diversity: Hardy–Weinberg expected heterozygosity (HHW), observed heterozygosity (HO), number of alleles per locus and the allelic richness (AR) per locus (to correct for biases in the number of alleles per locus associated with unequal sample size (Leberg, 2002)). We performed an analysis of variance to evaluate differences in HHW and AR among sampling sites, using locus as the block effect to remove interlocus variation. To evaluate the differences in genetic diversity (average HHW and AR) among marsh types, we conducted anova (PROC GLM, SAS Institute, 2005) using sampling sites as units of replication. These analyses allow us to evaluate the hypothesis that changes in the salinity gradient due to saltwater intrusion and the subsequent changes to marsh habitat could create barriers to movement, which coupled with survival challenges for many populations under higher salinity conditions could result in small isolated populations that would exhibit reduced allelic richness and lower genetic diversity relative to ‘freshwater’ populations.

To evaluate the spatial structuring of genetic variation across sampling locations, we calculated FST values (Weir & Cockerham, 1984) between and among all sampling locations using fstat 2.9.3 (Goudet, 1995). To examine the relationship among sampling locations, we used principal coordinate analysis, based on chord distance (Cavalli-Sforza & Edwards, 1967) to visualize the genetic distance between sampling sites both within and among marsh types. This analysis was conducted using genalex version 6.4 (Peakall & Smouse, 2006).

We evaluated the relationship between genetic and geographical distance to determine whether genetic similarity of sites followed an isolation-by-distance (IBD) model. To examine IBD, we defined genetic differentiation as FST/(1-FST) and evaluated the relationship between those values and the natural logarithm of the distance between sampling sites. Rousset (1997) indicated that an analysis using the logarithm of the geographical distance performed better for IBD in two-dimensional space. A matrix of rank habitat differentiation was created to compare genetic distance with the marsh types. A three-way Mantel test was conducted using xlstat (New York, NY, USA) to measure associations between genetic distance and marsh type while holding geographical distance constant, recognizing there is debate over whether the test produces biased estimates of P-values (Raufaste & Rousset, 2001; Castellano & Balletto, 2002; Rousset, 2002). Significance of associations was determined using 1000 randomizations of the distance matrix elements. Previous studies (Spencer et al., 1999, 2000) showed no indication of a violation of the neutrality assumption for our 8 microsatellite markers. However, to verify that assumption, we conducted a locus-by-locus amova as implemented in arlequin 3.5.1.3 (Excoffier & Lischer, 2010) to evaluate patterns of variation across all 8 loci.

We used two Bayesian assignment tests to examine the structure of populations. In the program structure 2.2 (Pritchard et al., 2000), we used the admixture model with correlated allele frequencies to evaluate a series of possible K values (1–15), where K is the possible number of populations among the sampled locations. At each of our estimated K values, we ran 5 independent runs consisting of 500 000 replicates with 50 000 burn-in replicates. stucture is one of the most common programs for the estimation of population structure; however, it has been reported to have reduced power under conditions of high gene flow (Latch et al., 2006; Waples & Gaggiotti, 2006; Chen et al., 2007). Due to the limitations of structure under conditions of high gene flow, we also analysed our data set using the admixture model implemented in tess 2.3.1 (Durand et al., 2009). The algorithm implemented in tess 2.3.1 has been shown to be a more efficient approach than structure because its iterative examination of K-values does not rely on additional information criterion values (Chen et al., 2007). We preformed analyses for K = 2–15 using the conditional autoregressive Gaussian model (CAR) with 100 replicate simulations for each value of K. Each simulation consisted of 50 000 sweeps with a burn-in of 30 000 sweeps, which included the use of a spatial prior to account for geographical relationships between collection localities. We visualized the lowest 10% of simulated runs based on deviance information criterion (DIC), and to determine the number of sampled clusters, we plotted the mean DIC values vs. K. We used the program clumpp 1.1.1 (Jakobsson & Rosenberg, 2007) to evaluate data for the most probable value of K, using 10 000 repeats of the Greedy algorithm to account for label switching, and these results were visualized using the program distruct (Rosenberg, 2004).

Results

Genetic structure within and between basins

Our analysis of genetic diversity between estuaries found no significant deviations from HWE or indications of LD in our microsatellite data set. We found an overall FST value of 0.039. Although all of the pairwise comparisons of genetic differentiation were significantly greater than zero, the levels of differentiation between estuaries were substantially larger than the differences within an estuary (Table 3).

Table 3. Genetic differentiation between sampled locations for G. affinis stock populations. Pairwise FST values are below the diagonal and P-values are above. Emboldened P-values were determined significant using a false discovery rate (FDR) adjusted for multiple comparisons to a level of 0.05 (Benjamini & Hochberg, 1995)
 HBHFSFSB
HB  < 0.001 < 0.001 < 0.001
HF0.009  < 0.001 < 0.001
SF0.0390.042  0.021
SB0.0650.0620.011 

Genetic structure along a salinity gradient

Our analysis of the genetic diversity along a salinity gradient within an estuary found no evidence for the presence of null alleles within the data set. Our tests for HWE and for the presence of LD found no significant deviations from expected values after a false discovery rate (FDR) adjustment of alpha (Benjamini & Hochberg, 1995). We found no significant differences in measures of AR (F2 = 0.62, = 0.538) and HHW (F2 = 1.91, = 0.151) among sampling sites. We also found no difference in the AR (F1 = 0.25, = 0.621) and HHW (F1 = 1.18, = 0.190) among marsh types (Table 2).

The overall FST for all sampled sites was 0.011 (Table 4). Of the 276 tests for pairwise differentiation conducted on our data set, 111 were significant based on alpha adjusted for multiple comparisons (Table 4). Many of these significant pairwise tests (60 out of 111) involved sites F6, F9 or I16. There was no consistent pattern of differentiation among sampling sites within a marsh type or among sites between marsh types.

Table 4. Genetic differentiation measured between fish collected at sample locations. Pairwise FST values are below the diagonal and P-values are above. Emboldened P-values are significantly differentiated at α = 0.05, adjusted for multiple comparisons (Benjamini & Hochberg, 1995)
6F1F2F3F4F5F6F7F8F9I10I11I12I13I14I15I16B17B18B19B20B21B22B23B24
F1 0.53600.38440.04480.1514 0.00010.6594 0.0116 0.00220.1163 0.0008 0.00280.1558 0.00530.0236 0.0002 0.0082 0.01870.0970 < 0.00010.0307 0.0066 0.00040.0364
F20.0192 0.32010.51070.28080.02810.88480.1861 0.00680.4313 0.01620.18900.44940.04350.6574 0.01860.37590.43010.06970.07860.16740.23910.10040.4277
F30.00170.0101 0.71240.0622 0.00020.13300.16210.03850.0430 0.0179 0.00180.3729 0.00940.1373 0.00010.21940.01990.0805 0.00080.06200.05090.1939 0.0115
F40.00000.01580.0004 0.1084 0.01170.11670.12240.02070.13070.0427 0.00200.1426 0.0082 0.01290.0501 0.01880.0878 0.00870.03420.02840.23460.11000.0253
F50.00210.01080.0001−0.0040  0.00170.53440.6984 0.00210.35390.05020.21670.73480.05170.0362 < 0.00010.10690.19060.18130.07370.02580.34450.04040.0300
F60.03380.00810.03470.03560.0251  0.0001 < 0.0001 0.0001 0.0004 < 0.0001 0.0001 0.0008 0.0005 0.0001 0.0001 0.0017 0.0001 0.0002 < 0.00010.0003 < 0.0001 < 0.0001 < 0.0001
F70.00490.00480.00780.00860.00350.0234 0.33710.04230.55150.00370.21190.3503 0.00010.3982 0.0003 0.01650.20060.0286 < 0.00010.10120.20290.03150.1453
F80.00510.02020.00190.00250.00370.04150.0084 0.03740.0679 0.0058 0.00860.4943 0.00490.1660 0.00050.64590.38880.1389 0.0003 0.01150.21050.05990.2718
F90.00800.00960.00310.00850.00630.02830.00390.0082  0.0092 < 0.0001 0.00310.0401 0.0004 0.0001 < 0.00010.0020 0.0016 0.0077 0.0003 < 0.0001 0.0009 < 0.0001 < 0.0001
I100.01090.01070.01010.00510.00790.03070.00860.01250.0107  0.01410.17180.80550.52340.65290.02270.23400.35970.09610.45570.36570.87110.46870.3139
I110.01010.01620.00480.00730.00530.03910.01140.01300.01260.0122  0.00040.3373 0.0028 0.0036 < 0.00010.03740.0387 0.0012 0.0002 0.00120.1310 0.0027 0.0055
I120.01800.00020.01620.01480.00640.01440.00250.01760.00870.00600.0119 0.0741 0.00040.0920 0.0001 0.0127 0.01430.1037 0.0008 0.00300.0310 0.0005 < 0.0001
I130.00680.01380.01210.00270.00500.03120.01340.01080.01280.00000.00600.0115 0.29460.7116 0.00200.08170.53510.07420.02470.08390.62570.17310.0684
I140.02370.01840.02130.01590.01650.03430.03090.02400.01930.00160.02080.02160.0113  0.0149 < 0.00010.08630.06570.05150.67370.17250.1780 0.0021 0.0045
I150.00960.00170.00470.00690.00420.02420.00030.00590.00910.00030.0101−0.00160.00500.0163 0.02120.53690.03440.25830.02530.11090.05890.07420.2782
I160.03520.01370.03320.02640.02940.02400.01800.04140.02480.02500.04050.01480.03410.03800.0181  0.0013 0.0001 0.0005 0.0001 < 0.0001 0.0031 0.0001 0.0001
B170.01060.01320.00440.00420.00620.03490.01350.00030.00970.00580.00680.00970.00710.01220.00070.0328 0.40260.85540.61110.06730.58490.23570.4116
B180.00070.00780.0042−0.00330.00000.02800.00060.00000.00330.00000.00100.0049−0.00200.01210.00050.02490.0000 0.40370.06710.12800.81150.15820.1584
B190.00620.01440.00730.00730.00300.03130.00890.00040.01110.00780.01050.00590.00860.01560.00000.03610.0000−0.0030  0.00610.11010.56820.02660.3995
B200.02960.01590.01620.01640.01650.04550.03060.02640.01860.00530.02070.02330.02060.00000.01830.04000.01170.01370.0211  0.00450.3213 0.0001 0.0090
B210.00940.00960.00810.00860.00320.01860.00730.01480.01420.00480.01090.00470.00950.01500.00380.02740.01090.00240.00590.0225 0.4023 0.00440.0423
B220.00690.01160.00440.00000.00030.03290.00590.00140.00710.00000.00520.00800.00200.00360.00200.02620.00000.00000.00000.00550.0029 0.42480.7238
B230.00550.01690.00250.00000.00000.03850.00720.00360.01410.00170.01150.01240.00530.01620.00230.03190.00340.00000.00620.01740.00860.0000 0.1318
B240.00770.01520.01170.00750.00560.03900.00390.00610.01690.00320.01640.01490.01390.02160.00190.03120.00710.00000.00230.02380.00690.00000.0015 

Subtle differences between sampling locations can be observed in the principal coordinate analysis (Fig. 2) based on chord distance (Dc). In this analysis, there was no relationship of PC1 with salinity (x axis, Fig. 2). There was however a relationship between PC2 and salinity (y axis, Fig. 2) with freshwater sites having greater value along the component. In addition, there was a weak relationship between PC3 (width of the circle) and salinity with freshwater sites having larger values in general. Together PC2 and PC3, both correlated with salinity, accounted for 35% of the variation in the distance matrix. We found no correlation between geographical distance (= 0.060, = 0.205) and marsh type (r = 0.036, = 0.273) with FST. There was also no significant correlation between marsh type and genetic distance (r = 0.004, = 0.474) when controlling for covariation by holding geographical distance constant. The locus-by-locus amova indicated that 98% of the molecular variation within our data set was within populations. Locus-specific variation ranged from 94.6–99.7 per cent for within populations and displayed similar trends for all sources of variation. No locus showed aberrant behaviour with respect to overall trends in variation across loci.

Figure 2.

Principal Coordinate Analysis plot of population divergences. The PC1 is x axis, PC2 is represented on the y axis and PC3 is represented by circle diameter. Divergence between sites was calculated based on a chord distance (Dc) matrix.

Results from our structure analysis indicated the presence of a single cluster within the sampled locations. We analysed our data set for a series of K maxima ranging from 1–15; the highest likelihood values were for a single cluster (K = 1) by a large margin. Further analysis conducted with tess 2.3.1, a program reported to be more sensitive than structure (Chen et al., 2007), showed a negative trend of DIC value relative to increasing number of K maxima. Visual inspection of cluster membership assignments showed no support for the presence of distinct clusters within our sample sites.

Discussion

Genetic structure within and between basins

Our previous study of the western mosquitofish in coastal marshes experiencing saltwater intrusion reported divergent selection resulting in genetically distinct phenotypes with differing tolerances to salinity (Purcell et al., 2008). Under a model of local adaptation, we would expect populations experiencing different salinity regimes within the same estuary would be more genetically similar for neutral markers than populations adapted to the same salinity regime across estuaries. This hypothesis was confirmed. There were significant differences in microsatellite variation between the fish from the two regions from which our breeding stocks were collected, and these differences were much larger than the differences between sites from different marsh types within each region. The observation that gene flow is greater between marsh types in the same region compared with the same marsh type in different regions indicates lower relative gene flow between fish from brackish marshes in Cameron and Terrebonne Parish than between different salinity marshes in the same region. Therefore, gene flow between brackish marshes in different regions may not be maintaining the adaptation to elevated salinities exhibited by these populations. Rather, the phenotypic differences (increased salinity tolerance) may be present as a result of the balance between the relatively higher gene flow between salinity regimes within an estuary and the strong selective forces of salinity stress on exposed populations.

Genetic structure along a salinity gradient

Local adaptation to differences in heterogeneous landscapes is generally expected to develop under conditions of limited gene flow between populations that are exposed to unique patterns of natural selection (Slatkin, 1987; Lenormand, 2002). Although there was significant differentiation among some pairs of sites, overall our results suggest most sites are connected via gene flow (overall FST = 0.011). Although FST is known to be biased downward for highly polymorphic loci (Jost, 2008), such as those used in this study, a measure designed to correct for this bias did not indicate high levels of differentiation (DEST = 0.021, results not shown). What differences there where in allele frequencies between sampling sites was not explained by geographical distance. In addition, we found some evidence that allele frequencies in the populations are weakly correlated with salinity gradients. However, there is no evidence of any discontinuities in frequencies, associated with the gradients, as would be expected if there was an absence of gene flow between populations in different salinity regimes.

However, the fish from various sampled sites do not seem to be part of a panmictic population because many of the pairwise comparisons conducted among sampling locations detected significant differences, indicating there are genetic differences among populations across the spatial landscape of the estuary. Moore et al., (2007) examined the effect of high gene flow on phenotypically diverged populations of Gasterosteus aculeatus and reported pairwise FST values between populations with high gene flow that were quite low with a mean of 0.012. They found that this level of differentiation among populations was sufficient to allow adaptive divergence and result in intermediate phenotypes.

The two assignment tests indicated a lack of population structure among individuals sampled across the salinity gradient in south-eastern Louisiana. However, the results of our principal coordinate analysis did indicate that there is spatial variation in the genetic data, and it is correlated with salinity exposure. This analysis, together with many significant pairwise FST estimates, indicates that gene flow is not high enough for the sites to be considered as one large panmictic population and suggests some limited structuring related to salinity.

Gene flow (Nm) is a function of the sizes of populations (N) and the number of migrants moving among them (m). It is unclear from the observed differences in allele frequencies the degree to which high N or high m is reducing differentiation, but the balance between drift, which increases with decreased population size and migration, has consequences for local adaption. The higher the effective sizes of populations, the lower the number of migrants that is necessary to minimize differences in gene frequencies. Large population sizes with small numbers of migrants in each generation would permit the development of divergent phenotypes in response to differing selective regimes. The limited drift experienced by large populations would favour the development of local adaptive divergences between populations (Kawecki & Ebert, 2004). However, as population size decreases, a higher degree of movement between sampled populations is required to prevent increased differentiation and would require increased selection differentials for differences in salinity tolerance to occur over short distances. The amount of individual dispersal between Gambusia populations in coastal marshes is unknown, but in other systems it is high (Brown, 1987). Coupled with the large reductions in Gambusia abundance observed in more-saline marsh types, relative to freshwater marsh (Hitch et al., 2011), and the higher degree of fragmentation of the former, it is likely that our inability to detect a loss of genetic diversity in brackish marsh samples may be due to at least moderate levels of genetic exchange.

The development of local adaptation in the presence of considerable gene flow has been seen before and could be more common than predicted by theory (Nosil, 2008). Gene flow, especially at low levels, may actually favour adaptation by facilitating the distribution of beneficial mutations, release from inbreeding depression or compensation for genetic drift (Garant et al., 2007). Gene flow could also be beneficial to diverging populations when a population responds to stress or environmental changes through the distribution of novel mutations. However, development of novel beneficial mutations is rare, and the adaptive response to changes in the environment is commonly derived from standing genetic variation within populations (Stockwell et al., 2003).

An alternative hypothesis to explain the observed patterns in the genetic data could be recent divergence between populations in different marsh types. If such divergence was recent, it could result in differentiation in phenotypes without causing changes in microsatellite alleles sufficient to change FST to an appreciable extent. FST is known to be influenced by both recent and ancestral gene flow (Neigel, 2002; Muir & Schlötterer, 2005). Although it is possible that the fish in different salinity regimes are recently diverged, none of the different assignment tests detected any evidence of that divergence. Furthermore, there are no obvious breaks in the distribution of G. affinis across the salinity gradient we studied; in fact, this species represents a considerable component to the biomass in this system (Hitch et al., 2011). In addition, the geomorphology of this region has existed in its current form since the development of the Lafourche lobe of the Mississippi delta complex (~1–2 kya) (Tye & Coleman, 1989; McBride et al., 2007). This means that while the salinity gradient in Terrebonne Parish marshes is currently undergoing positional shifts due to marsh alteration and changes in freshwater input, there has been and continues to exist a salinity gradient within these environments and therefore greatly complicates the hypothesis that these adaptations are a product of recent marsh-type divergences.

The occurrence of local adaptation in the presence of considerable gene flow is characteristic of populations subjected to a strong selective pressure. Hey (2006) reports that the genetic divergence of populations is often a product of natural selection operating differently in each population, promoting the success of hybrids at some genes while excluding others and therefore permitting genetic divergence in the presences of significant gene flow. In addition, Lenormand (2002) indicates that gene flow between locally adapted populations' results in a migration load that negatively affects fitness. In general, the interplay between gene flow and selective pressure is such that if migration is small relative to the strength of selection, adaptive divergence may occur. The selective agent in our system is salinity stress, and previous studies have shown that saltwater exposure can have strong negative effects on fitness and survival (Congdon, 1994; Kandl, 2001; Purcell et al., 2008). It is well established that the frequency of storm events and large storm-driven tides is increasing (Nicholls et al., 2007) and will continue to enhance saltwater intrusion into coastal marshes. Our results suggest that the adaptation for increased salinity tolerance reported by Purcell et al., (2008) represents an example of adaptive divergence in the face of relatively high gene flow. These findings illustrate the need for a deeper understanding of the evolutionary processes involved in shaping the adaptive responses of populations to changing environments. In addition, it will become increasingly important for managers to understand the adaptive potential and the evolutionary trajectory of populations in order to mitigate the consequences of continually changing environments.

Acknowledgments

This research was funded by the Louisiana Board of Regents and the United States Environmental Protection Agency's Science to Achieve Results (STAR) program. Although the research described in this article has been funded in part by the EPA's STAR program through grant R-82942001 to the Louisiana Board of Regents, it has not been subjected to any EPA review and therefore does not necessarily reflect the views of the Agency, and no official endorsement should be inferred. The project would not have been possible without the assistance of K. Barr, G. Athrey, H. Oliver, B. Adams and J. Purcell and the valued input of two anonymous reviewers. Part of this work was carried out using the resources of the Computational Biology Service Unit from Cornell University which is partially funded by Microsoft Corporation. This research complies with all legal requirements and was conducted under IACUC # 2004-8717-031.