Genetic Structure and the North American Postglacial Expansion of the Barnacle, Semibalanus balanoides

Abstract

Population genetic characteristics are shaped by the life-history traits of organisms and the geologic history of their habitat. This study provides a neutral framework for understanding the population dynamics and opportunities for selection in Semibalanus balanoides, a species that figures prominently in ecological and evolutionary studies in the Atlantic intertidal. We used mitochondrial DNA (mtDNA) control region (N = 131) and microsatellite markers (∼40 individuals/site/locus) to survey populations of the broadly dispersing acorn barnacle from 8 sites spanning 800 km of North American coast and 1 site in Europe. Patterns of mtDNA sequence evolution were consistent with larger population sizes in Europe and population expansion at the conclusion of the last ice age, approximately 20 000 years ago, in North America. A significant portion of mitochondrial diversity was partitioned between the continents (φ_ST = 0.281), but there was only weak structure observed from mtDNA within North America. Microsatellites showed significant structuring between the continents (F_ST = 0.021) as well as within North America (F_ST = 0.013). Isolation by distance in North America was largely driven by a split between populations south of Cape Cod and all others (P < 10⁻⁴). The glacial events responsible for generating allelic diversity at mtDNA and microsatellites may also be responsible for generating selectable variation at metabolic enzymes in S. balanoides.

Determining the genetic structure of populations is critical for understanding evolutionary and ecological processes, such as local adaptation and speciation. Knowledge of the life-history characteristics of an organism is thought to provide predictive power about the levels of gene flow and the correlated amount of genetic subdivision in that species (Loveless and Hamrick 1984; Bohonak 1999). For example, the ability to disperse over long distances is expected to result in little population subdivision because few effective migrants are necessary to homogenize genetic variation (Hartl and Clark 2007). In the marine realm, planktonic dispersers have the potential to travel great distances during larval stages, which should correspond with little genetic structure (Hedgecock 1986; Hellberg 1996; Deng and Hazel 2010). Nevertheless, some species that experience extended pelagic dispersal also show subdivision at small scales (Karl and Avise 1992; Young et al. 2002; Taylor and Hellberg 2003), and a recent review suggests that larval duration and genetic structure are, at most, weakly correlated (Weersing and Toonen 2009). When an obvious barrier to gene flow exists the reason for genetic subdivision is clear. In other cases subdivision can result from less predictable hydrodynamic dispersal patterns or nonrandom settlement (Grosberg and Cunningham 2001; Sotka et al. 2004; Véliz et al. 2006). Selection can also influence observed genetic structure and lead to estimates of significant population differentiation at very small spatial scales (Hilbish and Koehn 1985; Johannesson et al. 1995; Burton 1997; Schmidt and Rand 2001; Véliz et al. 2004).

How planktonic larvae connect patches of adult individuals is often unknown despite the necessity of understanding gene flow when assessing ecological roles and managing species (Palumbi 1994; Cowen et al. 2000). Estimates of genetic structure cannot always be translated directly into gene flow data because populations are rarely at migration–drift equilibrium (Whitlock and McCauley 1999; Duvernell et al. 2008; Hellberg 2009). Depending on population size and migration rates, patterns of genetic structure may represent contemporary levels of gene flow or reflect historical demography. However, estimates of genetic structure provide a necessary baseline to evaluate local adaptation in species and, when they are interpreted with caution, can be used to elucidate current dispersal patterns and population history (Duvernell et al. 2008).

In the western Atlantic, oscillating glaciations throughout the Pleistocene have resulted in the repeated expansion and contraction of species’ ranges as well as cross-oceanic recolonization and intermittent vicariance events that have prevented gene exchange between populations (Pielou 1991; Wares 2002; Hewitt 2004). These glacial effects may create recognizable population genetic patterns such as phylogeographic splits, signals of population expansion, and clines (Hewitt 1996, 2000; Baker et al. 2008; Maggs et al. 2008; Schmidt et al. 2008). In many species in the northern hemisphere populations have less genetic diversity at higher latitudes, a finding referred to as “southern richness to northern purity” (Bernatchez and Wilson 1998; Hewitt 2000; Hellberg 2009). Furthermore, for intertidal organisms, North American genetic diversity is a subset of that found in Europe due to more favorable habitat in glacial Europe compared with North America (Olsen et al. 2010). In a mitochondrial DNA (mtDNA) survey of 6 intertidal species from the North Atlantic, Wares and Cunningham (2001) found that 3 of the species surveyed were likely absent from the northwest Atlantic during the most recent glaciation, whereas the others likely persisted in North America south of the glacial advance or in other refugial habitats. In their study the acorn barnacle, Semibalanus balanoides, showed patterns of both persistence and recolonization in North America. Two distinct North American mtDNA clades were dated to 2 significantly different time periods, one preceding and one following the most recent glacial maximum approximately 20 000 years ago.

In the ecological genetics literature there has been some debate about how selection on metabolic enzymes appears to proceed in different ways over the North American range of S. balanoides (Holm and Bourget 1994; Schmidt and Rand 1999; Rand et al. 2002; Véliz et al. 2004; Flight et al. 2010). Recovery of the neutral structure of S. balanoides is critical because interpretations of species distributions can be confounded if fitness differences exist between local demes due to significant population substructure. Patterns of neutral variation in species provide the necessary context for interpreting studies of selection. Furthermore, the distribution of S. balanoides is used by ecologists to study, among other things, community interactions and predicted responses to climate change (Connell 1961; Wethey 1983, 1984; Bertness and Gaines 1993; Southward et al. 1995; Bertness et al. 2002; Mieszkowska et al. 2006; Altieri et al. 2007). Despite the need, there has been no analysis of neutral population structure using nuclear and mitochondrial markers in S. balanoides across the Atlantic (but see Flowerdew 1983; Dufresne et al. 2002).

The goal of the present study was to determine how the dynamic glacial history of the North Atlantic has influenced genetic subdivision and the distribution of genetic diversity in S. balanoides. The questions addressed include: 1) How is genetic diversity at nuclear and mitochondrial markers partitioned across the Atlantic Ocean in S. balanoides?; 2) At what scale within North American populations is significant substructure recovered, and is there a general pattern of isolation by distance?; 3) What do the data on both coasts suggest about the glacial history of the populations?; and 4) Can the demographic history of North American populations help explain patterns of selection previously observed in the system?

Materials and Methods

Sampling

Semibalanus balanoides specimens were collected from the rocky intertidal at 8 sites along the northwest Atlantic coast ranging from Rhode Island, USA to New Brunswick, Canada (Table 1) in the Spring and Summer of 2008. Individuals were removed from rock surfaces, placed on ice, and brought back to the laboratory where they were stored at −80 °C prior to DNA extraction. Additional individuals were sampled from Southwold, Suffolk, UK in the Spring of 2010, shipped in RNAlater (Ambion Inc., Foster City, CA) and stored at −20 °C. All individuals collected were adults, and new settlers were not sampled. The sampling scheme was designed to capture patterns of genetic isolation at multiple spatial scales. Two continents were sampled, and each of the 4 major regions in North America, abbreviated as Rhode Island (RI), Maine (ME), Saint John (SJ), and Miramichi (MR), was represented by 2 sites collected less than 30 km apart.

Table 1.

Sampling locations used in this analysis

Region	Site	Abbreviation	Latitude	Longitude
Rhode Island, USA	Captain R.W. Wheeler State Beach	RWRI	41°21'40'' N	71°29'18'' W
	Sakonnet Point	SKRI	41°27'51'' N	71°11'47'' W
Maine, USA	Small Point	SPME	43°42'36'' N	69°49'59'' W
	Pemaquid Beach	PBME	43°49'58'' N	69°30'49'' W
Saint John, CAN	Saint John Bay West	SWNB	45°15'13'' N	66°03'45'' W
	Saint John Bay East	SENB	45°15'40'' N	66°01'32'' W
Miramichi, CAN	Burnt Church	BCMR	47°11'34'' N	65°08'11'' W
	Neguac	NGMR	47°14'34'' N	65°04'31'' W
Suffolk, UK	Southwold	SWUK	52°20'02'' N	01°41'12'' E

Genotyping

For a portion of the samples, genomic DNA was extracted using a phenol–chloroform extraction following digestion of tissue with proteinase K (Sambrook et al. 1989). Additional samples were extracted by dissecting cirri and placing them in squish buffer (10 mM Tris, 1 mM EDTA, 25 mM NaCl) with proteinase K (Gloor and Engels 1991; Brown et al. 2001). Samples were heated at 37 °C for 2 h. Proteinase K was inactivated by increasing sample temperature to 100 °C for 4 min. DNA from the remaining samples was extracted with a Qiagen DNeasy extraction kit (Valencia, CA) following manufacturers instructions. Microsatellite amplifications improved with the cirral and Qiagen extractions.

The primers used for the PCR amplification of the control region of the mitochondria were described in Brown et al. (2001) (Iso2: 5′—TTACGGGCGTATTTTACTTG; 12S5′rev: 5′—AATACAACACGGACCTCAAC). PCRs for control region samples were done in 25 μl volumes with 2.5 μl of 10× buffer, 0.5 μl of 10 μM primer stocks, 2 μl of 5 μM dNTPs, and 1.5 mM Mg²⁺ with 0.5 to 1 units of Choice Taq (Denville Scientific). Thermal cycling conditions consisted of 5 min at 95 °C followed by 30–40 cycles of 95 °C for 30 s, 55 °C for 30 s, and 72 °C for 1 min. The final extension was 5–10 min at 72 °C. Following amplification, products were cleaned with a sodium acetate-ethanol precipitation and sequenced in one direction with an Applied Biosystems 3130xl Genetic Analyzer.

Microsatellite amplifications used primers for the dinucleotide loci Sebal13, Sebal14, and Sebal35 (Dufresne et al. 1999), and the tetranucleotide loci 3Gata and 4Gata that were originally developed by P.S.S. and used in Dufresne et al. (2002). The locus 2Gata, also originally developed by P.S.S., was also used (2GataF: 5′—AGACCGGGACCTACAGGCG; 2GataR: 5′—GGACCTCATCGCAGCACTCC). The loci 2Gata and Sebal13 could not be objectively scored in our analyses and were excluded from further tests. Conditions for microsatellite analysis were the same as described above for the control region amplification. However, we found increased success of microsatellite amplifications with the use of a 2.5 mM Mg²⁺ concentration, as opposed to the 1.5 mM concentration used for the control region. Fluorescently labeled microsatellite amplicons were run on Applied Biosystems 3130xl and 3100 Genetic Analyzers and scored in 2 triplexed sets for North American samples (set 1: Sebal13-HEX, Sebal14-FAM, 2Gata-NED; set 2: Sebal35-HEX, 4Gata-FAM, 3Gata-NED). European samples were run as 2 duplexed sets (4Gata-FAM, Sebal35-HEX; and Sebal14-FAM, 3Gata-NED). Microsatellites were scored using PeakScanner v 1.0 (Applied Biosystems).

Data Analysis

Mitochondrial DNA

Control region sequences were aligned in Sequencher v 4.5 (Gene Codes, Ann Arbor, Michigan) and BioEdit v 7.0.9.0 (Hall 1999). Homopolymer tracts and indels made some areas of the alignment subjective. As a result 2 alignments were made, one with all gaps and ambiguous regions removed and another with ambiguous regions and gaps included (Supporting Information). Unless otherwise stated. all analyses in this manuscript were conducted on the alignment with no gaps or regions of ambiguous alignment. All sequences have been deposited in GenBank (JF766799–JF766929). Sequences were exported and analyzed in DnaSP v 5.1 (Librado and Rozas 2009) and Arlequin v 3.5 (Excoffier et al. 2010). Basic measures of sequence variability including haplotype and nucleotide diversity (π) were assessed in each individual population and across all sites sampled. For North American populations, these parameters were regressed on latitude to determine patterns of diversity along the coast using R software v 2.12.1 (R Core Development Team 2008).

A phylogenetic tree of haplotypes was constructed in MrBayes v 3.1.2 (Huelsenbeck and Ronquist 2001; Ronquist and Huelsenbeck 2003) using a haplotype file generated from DnaSP v 5.1. MrBayes was run under a general time reversible model with the gamma parameter estimated for 1.5 × 10⁷ generations with sampling every 1000 generations. Of the 1.5 x 10⁴ sampled generations, 25% were discarded as a burn-in. Two separate runs of 4 chains at different heating levels (1.00, 0.83, 0.71, 0.62) were used. Convergence was assessed based on the stability of the standard deviation of split frequencies and the potential scale reduction factor. The posterior probability values for bipartitions on the braches of the majority-rule consensus tree were examined in FigTree v 1.3.1 (Rambaut 2007). An additional analysis of 1.5 × 10⁴ generations was run in MrBayes to confirm topological relationships. Support for consensus tree nodes based on 1000 bootstrap values from a likelihood analysis were generated with RAxML v. 7.0.3 (Stamatakis 2006) under a general time reversible model. To display the data, a statistical parsimony haplotype network (Templeton et al. 1992) was constructed using TCS v 1.21 (Clement et al. 2000) with a single uncertain base treated as missing. Analyses were run independently for samples from North America and from Europe using a parsimony threshold of 95%. When possible loops in the network were resolved according to the frequency and topology criteria of Pfenninger and Posada (2002).

To investigate population structure, mtDNA sequence evolution models were tested using AICc values in jModelTest v 0.1.1 (Guindon and Gascuel 2003; Posada 2008) based on the alignment with all sequences and gaps excluded. φ_ST estimates based on the best model were made in Arlequin v 3.5 and significance was assessed with 1000 permutations. Hierarchical analyses of molecular variance (AMOVAs) were also carried out in Arlequin using sequence distance for all samples and North American samples only. AMOVA was run for the unambiguous alignment and the alignment with ambiguous regions and gaps included (deletion weight = 1). Haplotype diversity, nucleotide diversity, and neutrality statistics, including Tajima’s D, Fu’s F_S and R₂, were calculated in DnaSP on the unambiguous alignment only (Tajima 1989; Fu 1997; Ramos-Onsins and Rozas 2002). Significance was assessed with coalescent simulations. One North American specimen was excluded from these analyses due to an uncertain base call (sample NGMR3; accession number JF766906).

Demographic history for S. balanoides was reconstructed using BEAST v 1.5.4 (Drummond et al. 2005; Drummond and Rambaut 2007). An HKY + Γ model, with a Bayesian skyline prior was run in duplicate for the North American and European samples separately, consistent with Wares and Cunningham (2001). Markov chain Monte Carlo (MCMC) analyses ran for 3 × 10⁷ generations with a burn-in of 3 × 10⁶ and a thinning interval of 3 × 10⁴. Each analysis started with a random tree and random seed and was run with a strict clock. Sequence evolution parameters were estimated by BEAST. Success of MCMC analyses was assessed based on the effective sampling size (ESS) and likelihood traces in TRACER v 1.5 (Drummond and Rambaut 2007). We chose the Bayesian skyline method because it does not rely on prior specification of a growth model (Pybus et al. 2000). The goal of this analysis was to find the time to most recent common ancestor (T_MRCA) for North American and European populations of S. balanoides under different assumptions regarding North American clades. Bayesian skyline plots, which estimate the product of the effective population size and the mutation rate through time along with credibility intervals, were also constructed for each continent. All these methods rely on selection of a mutation rate. Although, to our knowledge, no mtDNA control region mutation rate has been directly estimated from barnacles, we used the value of 19%/MY as estimated from another crustacean, penaeid shrimp (McMillen-Jackson and Bert 2003).

Microsatellites

Microsatellite data were formatted for use in Microsatellite Analyzer (MSA) v 4.05 (Dieringer and Schlotterer 2003) and exported for analysis in Arlequin and Genepop formats (Supplementary Table S1). Tests for linkage disequilibrium between loci were done using exact tests in Genepop v 4.0.5.3 (Raymond and Rousset 1995) (Supplementary Table S2). F_ST values for all pairwise comparisons were calculated in Arlequin, and significance was assessed with 1000 permutations. Tests for the presence and influence of null alleles were conducted using FreeNA (Chapuis and Estoup 2007) with the assumption of failed amplifications for loci that could not be objectively scored or had no signal (approximately 13% of amplifications). FreeNA returns F_ST values before and after a correction for null alleles. Estimates of F_ST/(1 − F_ST) before and after null allele correction were plotted as a function of pairwise great circle distance using R software and the formula for distance from Mills et al. (2007). Pairwise estimates of Jost’s D (Jost 2008) were made using the online application SMOGD (http://www.ngcrawford.com/django/jost/; Crawford 2010) based on the harmonic mean of the per locus estimates. AMOVA for microsatellites was done in Arlequin. The length of the flanking region was set to zero for each locus, but all analyses used allele identity and not repeat number, and are therefore F_ST, not R_ST, analogs. Allelic richness adjusted for sample size was calculated on a per site/per locus basis with MSA.

Results

Mitochondrial DNA

A total of approximately 265 high-quality base pairs were recovered from the control region, including gaps. The length was 237 bp after removal of gaps and ambiguous regions, resulting in 67 polymorphic sites (Supporting Information). Although the fragment length is shorter, the number of polymorphisms in the control region is greater than the 41 sites recovered in Atlantic populations by Wares and Cunningham (2001). Of the 131 sequences obtained there were 79 unique haplotypes including the one specimen with an uncertain base as unique. Haplotype diversity was greater in Europe than North America (Table 2), and there were no shared haplotypes between European and North American samples. Phylogenetic reconstruction of haplotypes left large polytomies given the relatively short fragment length and similarity of the sequences; however, the 50% majority-rule consensus tree indicates the presence of 2 predominantly North American clades in the data (Figure 1). Support for the clades was high based on the Bayesian analysis but marginal with likelihood (“A” posterior probability = 0.91, ML bootstrap = 55; “B” posterior probability = 1, ML bootstrap = 67; Figure 1). An additional run of 1.5 × 10⁷ generations in MrBayes recovered qualitatively the same results. The 95% limits of parsimony were reached after 6 steps for both the North American and European samples. There were 3 independent networks recovered from North America (labeled “A”, “B”, and “EL” on Figure 2) and 3 outlying haplotypes (labeled “X”). European samples resulted in a single network (labeled “E”) with 6 outlying haplotypes (labeled “Y”).

Table 2.

Summary statistics of mitochondrial diversity and neutrality tests

Site	Number	Nucleotide diversity	Haplotype diversity	Tajima’s D	Ramos-Onsins and Rozas’ R2	Fu’s FS
RWRI	11	0.0241	0.9455	−0.5414	0.1315	−2.374
SKRI	11	0.0319	0.9455	−0.1682	0.1322	−1.58
SPME	12	0.0314	0.9091	−0.1029	0.1439	−0.016
PBME	11	0.0342	0.9273	0.1493	0.1531	−0.239
SWNB	10	0.0268	0.9556	−0.0053	0.1551	−1.363
SENB	12	0.0286	0.9546	1.5739	0.2116	−1.317
BCMR	12	0.0278	0.8030	−0.0211	0.1567	1.794
NGMR	8	0.0124	0.4643	−1.5467	0.3013	2.477
North America	87	0.0293	0.8875	−0.8212	0.0803	−15.48 (0.001)
SWUK	43	0.0298	0.9967	−1.7776 (0.02)	0.0538 (0.012)	−41.41 (<0.001)
Total	130	0.0348	0.9490	−1.3714	0.0583	−70.52 (<0.001)

Significance was only estimated for sites pooled by continent and all sites. Values in bold were significant based on coalescent simulations in DnaSP. P-values are given in parentheses.

Figure 1.

Reconstruction of mtDNA haplotype phylogenetic tree. The observed topology is a midpoint rooted majority-rule consensus tree from MrBayes, with branch lengths in expected substitutions per site. Bayesian posterior probabilities are shown above the nodes. Values below the nodes correspond to bootstrap values from a likelihood analysis in RAxML. Filled circles at the branch tips correspond to North American haplotypes and hollow circles are European haplotypes. Numbers next to the circles are the number of occurrences of each haplotype in the data set. The 2 predominantly North American clades are labeled A and B.

Figure 2.

Statistical parsimony networks for North America (top) and European samples. Circles are proportional to haplotype frequency and branches indicate a single substitution. Small empty circles are unsampled haplotypes. In the North American clades, samples from south of Cape Cod are shown with gray and samples from north of Cape Cod are black. Haplotypes differing by greater than 6 substitutions could not be joined in networks from either continent at 95% confidence. A, B, and EL correspond to North American clades A and B, as well as European-like haplotypes sampled in North America. E shows the network of samples from Europe. X and Y are haplotypes from each continent that could not be connected at 95% confidence. Circles marked with a star indicate the haplotype most likely to be ancestral for each network (A = 25.7%, B = 24.0%; all EL equally likely). Network E has 2 such haplotypes because they were equally likely (14.75%).

Tests of sequence evolution were conducted with jModelTest and AICc values showed that the JC + Γ model to be the best fit for the data. The estimated transition/transversion ratio was 3.5505 and the Γ shape parameter was 0.184, and these values were specified in Arlequin. Significance of Jukes Cantor pairwise φ_ST values were based on 1000 permutations, and 11 of 36 (30.6%) pairwise comparisons were significant at α = 0.05 (Table 3). However, only 3 of 28 (10.7%) were significant within North America. There was no significant pattern of isolation by distance when pairwise φ_ST/(1 − φ_ST) was regressed on great circle distance between sites within North America (P = 0.4245). There was a significant relationship between distance and φ_ST/(1 − φ_ST) when the European population was included (P < 0.0001), but it was driven entirely by comparisons between the continents. AMOVA analysis showed approximately 28.1% of the variation was partitioned across the Atlantic (P < 0.0001), whereas only 3.2% was partitioned among populations within North America (P = 0.0899) (Table 4). AMOVA analyses on the alignment with gaps and ambiguous regions included yielded qualitatively the same result (27.6% of variation between continents, 2.4% within North America). The regressions of nucleotide and haplotype diversity on latitude were not significant for North America populations (P = 0.236; P = 0.090, respectively). However, there was a significantly higher proportion of individuals from south of Cape Cod in North American clade A relative to clade B (A = 14 of 44 from south of Cape Cod; B = 4 of 36 from south of Cape Cod; Fisher exact test P = 0.033).

Table 3.

Genetic structure and geographic distances between sampling locations

	RWRI	SKRI	SPME	PBME	SWNB	SENB	BCMR	NGMR	SWUK
RWRI		−0.0258	0.1370	0.0343	0.3064	0.1271	0.2202	−0.0323	0.3011
SKRI	26.9		−0.0070	−0.0569	0.1299	0.0049	0.0517	−0.0351	0.2429
SPME	294.3	273.5		−0.0421	−0.0227	−0.0406	−0.0684	0.0759	0.2915
PBME	318.8	297.2	29.1		0.0595	−0.0196	−0.0013	0.0040	0.2334
SWNB	616.2	591.2	344.8	315.8		−0.0008	−0.0387	0.2653	0.3920
SENB	618.9	593.9	347.8	318.8	3.0		0.0038	0.0712	0.3362
BCMR	821.7	798.3	532.9	505.7	227.1	225.4		0.1648	0.3149
NGMR	829.0	805.5	540.1	512.9	233.8	232.1	7.2		0.3533
SWUK	5450.4	5424.3	5182.4	5153.7	4839.6	4836.7	4660.5	4653.6
RWRI		−0.0103	0.0178	0.0117	0.0160	0.0279	0.0337	0.0171	0.0344
SKRI	−0.0049		0.0169	0.0196	0.0143	0.0240	0.0274	0.0236	0.0443
SPME	0.0156	0.0151		0.0021	0.0055	0.0018	0.0050	0.0001	0.0261
PBME	0.0177	0.0182	0.0007		0.0030	0.0012	0.0145	0.0192	0.0223
SWNB	0.0148	0.0167	0.0064	0.0074		0.0064	0.0010	0.0005	0.0110
SENB	0.0270	0.0250	0.0012	0.0041	0.0051		−0.0002	0.0038	0.0151
BCMR	0.0319	0.0300	0.0056	0.0191	0.0067	0.0034		−0.0019	0.0366
NGMR	0.0242	0.0218	0.0034	0.0182	0.0073	0.0067	0.0080		0.0294
SWUK	0.0463	0.0476	0.0304	0.0262	0.0213	0.0216	0.0418	0.0273

The top matrix shows pairwise mtDNA φST estimates based on Ti/ Tv and Γ corrected Jukes Cantor distances above the diagonal and great circle distance between sampled sites in kilometers below the diagonal. The lower matrix shows pairwise microsatellite F_ST values estimated before correction for null alleles above the diagonal and after correction below. Values above the diagonals in bold show significant differences based on permutation tests. Values calculated after null allele correction are not labeled as significant because 5 loci are required for a significance estimate in FreeNA software.

Table 4.

AMOVA analyses of mtDNA and microsatellite data

Level of structure	Percentage of variation	Fixation index	P value
mtDNA
Among continents	28.14	FCT = 0.28145	<0.00001
Among sites in North America	3.2	FSC = 0.04451	0.08993
Within sites	68.66	FST = 0.31343	<0.00001
Microsatellites (all samples)
Among continents	2.13	FCT = 0.02132	0.00293
Among sites in North America	1.27	FSC = 0.01295	<0.00001
Among individuals within sites	9.77	FIS = 0.10118	<0.00001
Within individuals	89.83	FIT = 0.13173	<0.00001
Microsatellites (North America only)
Among regions	1.34	FCT = 0.01338	0.00098
Among sites in regions	0.17	FSC = 0.00175	0.19355
Among individuals within sites	9.36	FIS = 0.09499	<0.00001
Within individuals	89.13	FIT = 0.10866	<0.00001

These mtDNA data are from the unambiguous alignment with no gaps. The gapped alignment gave qualitatively the same result (see Results section).

Of the neutrality statistics tested in both populations, D and R₂ were significantly negative for the European samples only, and Fu’s F_S was significantly negative for both populations (Table 2). Demographic reconstructions in BEAST had ESS values above 100, usually greatly so, for all parameters (Rambaut and Drummond 2007). Bayesian skyline reconstructions show a rapid recent increase in population size for the North American population (Figure 3). T_MRCA for the North American lineages was estimated to be 1.90 × 10⁵ (95% CI: 1.11–2.99 × 10⁵ years). The European lineages show slower and older population expansion but a similar T_MRCA to that observed in North America (median = 1.78 × 10⁵ years; 95% CI: 1.06–2.82 × 10⁵ years).

Figure 3.

Bayesian skyline plots (top) and T_MRCA estimated for North American and European barnacle populations. Solid lines in the skyline plot represent North American samples and dashed lines European samples. The center lines are the median log₁₀ of N_e times the mutation rate as a function of thousands of years before present (estimated assuming a substitution rate of 19%/My). Outer lines are the 95% highest posterior density. The lower graph shows the frequencies of estimated T_MRCA for North American clades A and B, and the continental populations for Europe, “EUR,” and North America, “NA.” On both graphs, an estimate of the last glacial maximum (∼20 000 years before present) is shown with a vertical gray line.

Microsatellites

Of the 4 genotyped loci, potential null allele frequencies were variable across loci and site (Table 5), but estimates of F_ST after null allele correction were similar to uncorrected estimates made in Arlequin (Table 3). Overall, 21 of 36 (58.3%) pairwise F_ST values were significant at α = 0.05 based on 1000 permutations in Arlequin (Table 3). Regression of F_ST/(1 − F_ST) on pairwise great circle distances between sites showed a highly significant pattern of isolation by distance before (P < 0.0001) and after (P < 0.0001) correction for null alleles within North America (Figure 4). Isolation by distance was also significant across the Atlantic, but the regression was largely influenced by the split between the 2 continents. AMOVA analysis partitioned 2.13% of the variation in microsatellite markers across the Atlantic, as opposed to 1.27% among the sites in North America (Table 4). Allelic richness was greatest in the European population for 3 of 4 loci (Table 5), but there was no significant relationship between allelic richness and latitude in North America for any locus (P > 0.05 in all cases).

Table 5.

Summary statistics for microsatellite data

	Sebal14					Sebal35					3Gata					4Gata
Site	N	A	AR	Hobs	Null	N	A	AR	Hobs	Null	N	A	AR	Hobs	Null	N	A	AR	Hobs	Null
RWRI	35	10	9.9	0.857	0.000	36	15	15.0	0.889	0.013	32	10	10.0	0.813	0.040	41	20	19.8	0.902	0.000
SKRI	43	8	7.9	0.721	0.005	42	12	11.8	0.881	0.000	41	10	9.7	0.732	0.049	40	22	21.9	0.875	0.034
SPME	38	10	9.9	0.737	0.048	40	13	12.4	0.825	0.000	44	8	7.2	0.477	0.112	42	20	19.7	0.857	0.018
PBME	42	9	8.4	0.690	0.057	38	17	16.6	0.816	0.005	45	8	7.4	0.533	0.111	39	17	17.0	0.795	0.053
SWNB	40	7	6.8	0.800	0.028	36	16	16.0	0.750	0.067	32	10	10.0	0.531	0.134	40	21	20.8	0.775	0.072
SENB	41	11	10.6	0.659	0.082	39	8	7.9	0.744	0.050	37	12	11.4	0.649	0.076	43	16	15.6	0.884	0.000
BCMR	34	6	6.0	0.647	0.046	42	12	11.4	0.714	0.040	42	8	7.4	0.476	0.096	43	18	17.8	0.930	0.004
NGMR	41	10	9.6	0.659	0.035	40	12	11.4	0.775	0.010	40	9	8.3	0.525	0.107	40	18	17.9	0.850	0.000
SWUK	42	14	13.1	0.810	0.000	47	15	13.5	0.766	0.059	41	15	13.7	0.415	0.238	45	23	22.3	0.978	0.000

The number of individuals sampled (N), the number of different alleles (A), allelic richness (AR), observed heterozygosity (H_obs), and estimated frequency of null alleles are given.

Figure 4.

Isolation by distance based on microsatellite loci. The large graph corresponds to sites in North America only. Triangles are comparisons that include populations south of Cape Cod and circles are northern populations only. The inset graph shows all comparisons, including those across the Atlantic. Filled symbols (solid lines) are estimates from Arlequin and hollow symbols are estimates from FreeNA (dashed lines) after correction for null alleles. All trendlines were significant at P < 0.0001.

Discussion

Genetic Structure

Although both sides of the Atlantic have undergone glacial events for hundreds of millennia, the influence of glacial advance and recession appears to differ on the 2 continents. One general pattern that has emerged is that glacial events in the last 100 ky have been more severe in the western Atlantic. This results in several predictions, including lower genetic diversity in the western Atlantic and evidence for recolonization of North America from Europe. Previous analyses of mtDNA structure in S. balanoides have revealed that North American mtDNA can be divided into 2 haplogroups (Brown et al. 2001; Wares and Cunningham 2001). The phylogenetic and parsimony analyses here support this result, with high Bayesian posterior probabilities for 2 predominantly North American groups. Some European haplotypes do nest within clade B in the consensus tree (Figure 1), but this is consistent with European populations being ancestral to North American populations. Diversity is greater at the haplotype and nucleotide levels in the European population than in the North American samples, and there are no shared haplotypes between the continents. This lack of shared haplotypes indicates that the effective mutation rate is likely higher than the cross-Atlantic migration rate. AMOVA estimates partition 28.1% of the variance to the cross-oceanic split. Within North American populations there was no evidence for significant isolation by distance for mtDNA using φ_ST estimates.

As with mtDNA, analysis of the microsatellite data suggested higher allelic richness in Europe and significant structure between North American and European populations. However, F_ST estimates across the Atlantic are low, which is likely a function of the high level of polymorphism observed in the microsatellites and potential homoplasy of alleles (Hedrick 1999; O'Reilly et al. 2004). A separate analysis of the microsatellites using Jost’s D is available as Supporting Information (Supplementary Table S3). Within North America, the microsatellite data indicate that despite large population sizes and the potential for frequent long-distance migration, barnacle populations exhibit a highly significant pattern of isolation by distance. The low values of F_ST are consistent with long distance dispersal and high heterozygosity, but comparisons of all sampling localities reject panmixis in the northwest Atlantic. Barnacle nauplii spend 5–8 weeks in the plankton and can potentially disperse hundreds of kilometers in that time (Drouin et al. 2002). With this dispersal capability, panmixis may have been a likely result. A significant portion of the variation in microsatellite data is also partitioned among individuals within sites, a result previously seen in marine invertebrates and termed “chaotic patchiness” (Johnson and Black 1982). In S. balanoides, this effect may be due to the sampling of patches of related individuals within sites (Véliz et al. 2006).

Even when isolation by distance is observed, populations may not be in migration–drift equilibrium (Duvernell et al. 2008). In these cases, historical isolation of populations may also result in isolation by distance that does not reflect current dispersal patterns. In some species, Cape Cod presents a genetic breakpoint, indicative of a barrier to dispersal or historical isolation (Wares 2002; Jennings et al. 2009). In this analysis, 12 of 14 significant pairwise F_ST values in North America include a population south of Cape Cod, and the pattern of isolation by distance is not significant when these comparisons are excluded from the linear regression (P = 0.076; Figure 4). This suggests that historical effects associated with Cape Cod, as is the case in other species, may be more responsible for the observed structure than limited gene flow. Two of the 3 significant φ_ST comparisons within North America also include populations from south of Cape Cod (Table 3), and haplotype frequencies are significantly different across the peninsula. The differentiation of populations south of Cape Cod relative to the rest of North America could be indicative of expansion from an ancient southern refugium.

Genetic Diversity and Demographic History

Neutrality tests based on the frequency spectrum of mutations (Tajima’s D, Ramos-Onsins and Rozas’ R₂) and haplotype distributions (Fu’s F_S) were consistently indicative of demographic expansion or purifying selection for the European samples (Ramírez-Soriano et al. 2008). Within North America and among all samples, only Fu’s F_S was outside the 95% confidence intervals based on coalescent simulation. Fu’s F_S is generally a more powerful test of population expansion than tests based on the site frequency spectrum if there is no recombination (Ramírez-Soriano et al. 2008). Given that these data come from the nonrecombining mtDNA, Fu’s F_S may have had increased power to detect expansion in all samples. However, homoplasy due to hypervariable sites within the control region could also increase the number of haplotypes relative to the number of segregating sites, without strongly influencing site frequency statistics.

Bayesian skyline plot analyses are indicative of differing population histories on the 2 Atlantic coasts. Within North America, genetic diversity is generally lower and shows a recent expansion beginning approximately 20 000 years ago. Although this figure is dependent on the selection of a mutation rate, the observed result is in close agreement with the last glacial maximum. The mutation rate of 19%/My is high as compared with the 1.29%/My estimated for Chthamalus 16S mtDNA sequences by Wares et al. (2009). However, substitution rates are drastically elevated in the hypervariable mtDNA control region across a range of taxa. In the penaeid shrimp described in the Materials and Methods section, COI mutation rates were estimated at 1.5%/My, suggesting that the control region has greater than 10-fold higher substitution rate than a coding region of the mtDNA (Baldwin et al. 1998), most likely due to differences in purifying selection (Haney et al. 2010). By contrast, the European population shows greater diversity and a signature of expansion predating that in North America.

Previous coalescent analyses in this species indicate a more recent demographic expansion in North America (Wares and Cunningham 2001), which is common for species following glacial retreat (Hewitt 2004; Maggs et al. 2008). This result is also observed in these data. Wares and Cunningham’s (2001) analyses also divided the North American population of S. balanoides into 2 clades, which were analyzed independently to determine the time at which the population size was one percent of the current value. The main finding of Wares and Cunningham (2001) regarding S. balanoides is that they identify 2 significantly different coalescent ages for the 2 clades that bracket the last glacial maximum. In this analysis, the 2 North American clades have different estimated coalescent times but broadly overlapping confidence intervals (Clade B: 3.70 × 10⁴ years, 95% CI 1.80–6.84 × 10⁴; Clade A: 4.96 × 10⁴, 95% CI 1.81–10.3 × 10⁴; Figure 3). Clade B has a very similar T_MRCA as the “North” clade reported in Wares and Cunningham (2001) (∼33 000). However, clade A appears to be younger than previously estimated. One important difference between the studies is that Wares and Cunningham (2001) estimated time until the population was 1% of current size. Under an exponential growth model, as they assumed, this is very different from the T_MRCA. In the case of the North clade, they report a T_MRCA approximately twice as great as time to 1% of current size. Therefore, although the “South” clade in their study shows a time to 1% of current size of 68 000 years, the T_MRCA is expected to be a great deal older (∼136 000 if we assume doubling). This estimate is more than twice our estimate of ∼50 000 years T_MRCA for clade A.

The discrepancy in coalescence times between studies may be due to the inclusion of individuals with “European-like” haplotypes into the clades sampled by Wares and Cunningham (2001). When these individuals are included as part of clade A in these data, the estimate for T_MRCA is much greater (1.72 × 10⁵, 95% CI 9.39–27.5 × 10⁴). However, given the strong support for reciprocal monophyly of clades A and B in the phylogenetic and network analyses (Figures 1 and Figure 2), we did not feel justified including these haplotypes (labeled EL and X in the Figure 2) as members of either A or B. Furthermore, the average number of pairwise substitutions between sequences in the North American EL network and the E network from Europe (7.4 substitutions) was fewer than the average number of substitutions between EL and A (10.3) and between EL and B (9.1). One possibility is that these European-like outliers (∼9% of the North American data in Figure 2) may represent recent colonization events of North America from Europe, potentially as a result of ship traffic. Barnacles have a long planktonic dispersal and could potentially cross the Atlantic in ballast water. Moreover, barnacles are well known as biofoulers that can travel with ocean going vessels (Southward et al. 1998; Yamaguchi et al. 2009). Under these circumstances, these individuals could represent European haplotypes sampled in North America due to recent transport. These haplotypes remain at low frequencies compared with the clades that have likely persisted in North America.

Taken together, the evidence provided here clearly shows postglacial expansion and significant population substructure, but the origins of diversity in S. balanoides have multiple possible explanations. It could be, as suggested by Wares and Cunningham (2001), that North American S. balanoides represent 2 distinct colonization events that occurred on opposite sides of the most recent glacial maximum. However, we cannot rule out a single colonization event with ancestral polymorphism in the initial migrants followed by simultaneous expansion of each clade. Wares and Cunningham (2001) rejected this scenario due to the discrepancy in ages of the 2 North American clades, a result not observed when potential recent colonizers are excluded. Other explanations are also possible. Contraction of the population during glaciation could have eliminated lineages and led to deeper coalescence of the clades that survived the bottleneck. Finally, the clades could have originated in independent glacial refugia in North America. These scenarios are difficult to distinguish with current data due to the stochastic nature of the coalescent process. The greater allelic diversity on both genomes in Europe supports the finding of Wares and Cunningham (2001), that European populations are ancestral. However, it is not necessarily true that the T_MRCA reflects the age of the North American populations. Rather, it describes the time since coalescence of the sampled North American lineages. The geographic location of this event and the population parameters prior to the coalescence of these specific lineages are unknown. Resolution of the number of colonization events and their temporal sequence will require sampling of more loci to increase the number of gene trees and the power of the analysis.

A Note on Selection

The idea that populations of S. balanoides from North America and Europe may be distinct races dates to at least the 1960s (Crisp 1964). Flowerdew (1983) proposed that different allele frequencies at 2 allozyme loci, Mpi and Gpi, across the Atlantic supported the hypothesis of different races. The loci are primarily 2 allele systems in North America with common and rare alleles at each locus segregating at approximately 70% and 30%, respectively. However, Holm and Bourget (1994) later rejected the idea that Mpi and Gpi provide evidence for racial differences because they found that allele frequencies at these loci vary over a wider range than observed by Flowerdew (1983) on small scales within North America, likely as a result of selection. Subsequent work (Schmidt and Rand 1999; Véliz et al. 2004) further supported the idea of selection at these loci, but the exact mechanisms and scale of selection remain controversial (Rand et al. 2002; Flight et al. 2010). Dufresne et al. (2002) previously noted a potential link between selection at metabolic enzymes in this species and allele frequencies at the Sebal13 and Sebal14 microsatellite loci. In this study, Sebal14 does show significant isolation by distance, but F_ST remains low across more than 800 km of coastline.

The origin of nonneutral alleles at Mpi and Gpi may stem from the same glacial population dynamics that led to the current distribution of mtDNA diversity. Flowerdew (1983) observed more alleles at nuclear loci in Europe than in North America. This result has further been observed for mtDNA by Wares and Cunningham (2001) and, in this study, for mtDNA and microsatellites. When the evolution of allelic forms has been influenced by metapopulation dynamics extra caution must be taken with traditional tests of selection (Rand 1996; Jensen et al. 2005). We have shown here that mtDNA diversity has been strongly influenced by past glacial cycles. The origin of selectable variation at the Mpi and Gpi loci in S. balanoides may have been generated by the effects of glaciation rather than through de novo mutation in a panmictic population. The scenarios for the origins of diversity described here including separate colonization events, ancestral polymorphism, separate isolated refugia, or bottlenecks all have the potential to produce 2 allele systems with deep coalescences between allelic forms. Identifying which scenario is specifically responsible for the origin of diversity in S. balanoides will be difficult, but could provide an example of how glacial events may produce selectable variation in current populations. The fact that populations south of Cape Cod are the primary drivers of isolation by distance in North America suggests that the different selective patterns observed across more northern populations are the result of different environmental pressures rather than fixed differences in alleles. It is possible that patterns of F_ST at the Mpi and Gpi loci may depart from the global patterns inferred here from mtDNA and microsatellites, but this would also be entirely consistent with fitness variation at those loci.

Conclusions

In this study we set out to answer several questions with regard to how genetic diversity in S. balanoides is partitioned, both across the Atlantic and within North America. These questions are framed by the glacial history of both coasts of the Atlantic, and the previous observation that North American populations of S. balanoides may represent 2 colonization events. We find strong evidence that North American and European populations are significantly different from one another using both mtDNA and microsatellite markers. Microsatellites also show isolation by distance within North America, largely driven by a split between the southernmost populations and all others. As predicted based on the severity of glaciation 20 000 years ago on both coasts, the population in North America is less genetically diverse and has undergone more recent growth. As for the origins of the current North American population, we cannot rule out the previous conclusion of Wares and Cunningham (2001), that the 2 North American haplogroups represent separate colonization events. However, other explanations for the origins of diversity in North America remain possible, including multiple refugia in North America, a North American bottleneck event, or ancestral polymorphism in a single colonization event.

Supplementary Material

Supplementary material can be found at http://www.jhered.oxfordjournals.org/.

Funding

National Science Foundation (grant DEB-0108500 to D.M.R.); Experimental Program to Stimulate Competitive Research (EPSCoR) Graduate Research Fellowship (to P.A.F.); Summer Undergraduate Research Fellowship from the Rhode Island EPSCoR program (to M.A.O.); National Institutes of Health grant GM067862 (to D.M.R.).