Phylogeography of postglacial range expansion in Nigronia serricornis Say (Megaloptera: Corydalidae)

Geography and collection of sample populations

To obtain a comprehensive understanding of the underlying historical population structure of Nigronia serricornis, 344 individual larvae representing 30 populations were sampled from across the range of the species (Fig. 1). The larvae of N. serricornis exhibit unambiguous morphological differences from the only other species in the genus, Nigronia fasciatus, allowing specific identification in the field. At each site, up to 18 N. serricornis larvae were collected by kick-netting and immediately preserved either on dry ice or in liquid nitrogen. Larvae were stored at −80 °C until DNA could be extracted. Whenever possible, sampling sites were selected where N. serricornis had recently (= 5 years) been collected by academic researchers or governmental bio-assessment units. One Corydalus cornutus (Vermillion County, Illinois) and four N. fasciatus larvae (representing 2 sites in West Virginia and 1 in Indiana) were collected and used as outgroup taxa (geographic sampling information available from the authors upon request). Voucher specimens for all Nigronia populations analysed were deposited in the insect collection of the Illinois Natural History Survey (INHS catalogue nos 39226 –39569 and 39570–39573, for N. serricornis and N. fasciatus, respectively).

MtDNA data collection

Initially, DNA was extracted from head, thoracic, and abdominal preparations. Extractions from the larval head capsule were used for later amplification, because they resulted in the cleanest extractions. All DNA extractions were performed using QIAGEN DNeasy™ kits, following the manufacturer's protocol for animal tissue samples. To offset the effects of an unknown polymerase chain reaction (PCR) inhibitor that co-eluted with the DNA, DNA elutions were diluted 1:150 with ultrapure water to ensure reliable amplification.

For all specimens, a 710-base fragment in a coding region of cytochrome oxidase subunit I (COI), previously shown to be useful in answering population-level questions (Wishart & Hughes 2003), was amplified using the primers LCO1490 (5′-GGTCAACAAATCATAAAGATATTGG-3′) and HCO2198 (5′-TAAACTTCAGGGTGACCAAAAAATCA-3′) of Folmer et al. (1994). Amplification reactions were prepared using the following protocol: 10.3 µL double processed tissue culture water (Sigma), 2.5 µL PCR Gold buffer (ABI), 1.0 µL 25 mm MgCl₂, 1.0 µL 5 mm dNTPs, 1.0 µL each 20 mm primer, 0.2 µL AmpliTaq® Gold enzyme, and 8 µL diluted template. Amplification reactions were performed on an MJR PTC100, following the thermocycler protocol of Wishart & Hughes (2003). Amplification reactions were electrophoresed on 1.5% agarose gels for 45 min at 120 V, stained with ethidium bromide, and gels were visualized using UV light. PCR products were purified using QIAGEN QuickSpin™ kits as per the manufacturer's specifications.

The following sequencing reaction was prepared using the purified DNA: 5.2 µL 12.5% glycerin, 2 µL 5X sequencing buffer (ABI), 2 µL 20 mm PCR primer, 1.0 µL BigDye terminator (version 3.0, Applied Biosystems), and 6–8 µL purified DNA and subjected to 35 cycles of PCR using the conditions of Collins (2003). DNA fragments were sequenced in both directions. DNA sequencing was performed on an ABI 3730X1 by the University of Illinois Biotechnology Center.

Sequence verification and Bayesian phylogeny

Sequence chromatograms were verified and sequences were manually aligned, as no gaps were present in the sequences. All DNA sequences were deposited in GenBank (accession nos AY750172–AY750519). A 630-base fragment of COI was used in subsequent analyses. To ensure that PCR products were from true mitochondrial genes, a nucleotide-nucleotide blast search was performed on the most common haplotype. The corrected pairwise COI sequence divergence was calculated in paup* (version 4.0b10, Swofford 2003) using the same model as the Bayesian analysis (see below). A Bayesian phylogeny was constructed for unique COI haplotypes to identify ancestral haplotypes with the program mrbayes (version 3.0, Ronquist & Huelsenbeck 2003) using 2 000 000 trees, sampling 1/100 trees, with a burn-in value of 837, corresponding to generation 83 600. C. cornutus served as the outgroup taxon. The program modeltest (version 3.5, Posada & Crandall 1998) was used to determine the appropriate substitution model for COI (model = TVM + I + G). The burn-in value was selected by plotting the likelihood values a posteriori and selecting a generation beyond the point of stationarity (Ronquist et al. 2005).

Population genetic analyses

Standard population genetic and statistical analyses were employed to describe the main features of the data. In some of the population analyses, only samples with n = 8 (henceforth the ‘large’ samples) were included; a compromise between the desire to maximize samples, while including as much of the geographic range as possible (mean n of large populations = 16). As a measure of whether the data were consistent with selective neutrality, Tajima's D was calculated using arlequin 3.0 (Excoffier et al. 2005) for both the entire set of sequences considered together and for the individual large populations. The overall magnitude of population differentiation was measured by calculating F_ST across ungrouped samples with haplotype frequency data using arlequin (Excoffier et al. 2005). As an overall assessment of whether the population differentiation showed geographic structure, a comparison of the two fixation indices, G_ST (Nei 1982) and N_ST (Lynch & Crease 1990) was carried out using dnasp (version 4.0, Rozas & Rozas 1995). If the value of N_ST (analogous to G_ST, except that sequence differences as well as frequency differences between alleles are incorporated in the statistic) is larger than the value of G_ST for a given set of data, the presence of phylogeographic structure can be assumed (Pons & Petit 1996). Finally, the relationship between latitude and both the proportion of haplotypes at a sampling site (# haplotypes/n) and haplotype diversity were examined by linear regression, after dnasp was used to determine the haplotype diversity.

Phylogeographic analysis

Two methods, NCA and analysis of molecular variation (amova; Excoffier et al. 1992), were employed to examine the relationship between intraspecific phylogeny and geography. To implement the NCA, an explicit parsimony (Templeton et al. 1992) haplotype network was first constructed for the COI sequences using the program tcs (version 1.18, Clement et al. 2000), with connections made at the 95% parsimony limit (10 steps). The few ambiguous connections in the resulting haplotype network were resolved as per Crandall & Templeton (1993). The haplotype network was then overlain upon geographic space to look for gross patterns in the data, prior to being partitioned into nesting clades as per Templeton (1998). Information regarding clades that contained appropriate amounts of genetic and geographic variation was input into geodis (version 2.0, Posada et al. 2000) together with geographic sampling information, using the latitude/longitude method, to perform permutation tests for association between phylogeny and geographic distribution. Templeton's revised (2004) inference key was then applied to the results from geodis to determine the outcome of the NCA.

An amova was performed on populations grouped phylogenetically (groupings available upon request of the authors) using arlequin 3.0 (Excoffier et al. 2005). A global amova was carried out on the large population samples, using the following groupings of populations suggested by the NCA: clade II-19, Tennessee clade, Coastal clade, North Carolina clade, Georgia clade, and central Kentucky clade (see Fig. 1).

Sequence variation and Bayesian phylogeny

The 630-base fragment of COI (containing 58 polymorphic and 38 parsimony-informative sites) was amplified and sequenced for 344 Nigronia serricornis larvae. Of these sequences, 68 unique haplotypes were recovered (see Table 1 for geographic occurrences). Approximately 94% of the polymorphic sites represented changes at third codon positions, and all substitutions were synonymous.

No stop codons were present in the mtDNA sequences, indicating that a true mitochondrial gene was amplified, as opposed to a nuclear pseudogene. The nucleotide-nucleotide blast search recovered a number of significant matches for COI sequences from other insects with over 80% sequence identity. Nigronia serricornis was different from Nigronia fasciatus by only a single amino acid, and the N. fasciatus sequences formed a monophyletic sister group to N. serricornis. Sequence divergence within N. serricornis ranged between 0 and 0.044 substitutions/site, with a mean divergence of 0.020 substitutions/site. The level of divergence between N. serricornis and N. fasciatus ranged between 0.195 and 0.245 substitutions/site, with a mean sequence divergence of 0.219 substitutions/site.

The Bayesian tree confirmed the monophyly of N. serricornis with respect to N. fasciatus, and posterior probabilities were greater than 90% for a majority of uncollapsed nodes (Fig. 2). The phylogeny rooted with Corydalus cornutus showed a group of haplotypes concentrated in Pennsylvania and New York (haplotypes 53–59; Fig. 2) to be ancestral for N. serricornis, with an overall ln(likelihood) of 2602.41.

Population genetic analyses

Nucleotide diversity over the entire data set was 0.015 ± 0.008. The value of Tajima's D using all 344 sequences was −0.17524, not significantly different from 0 (P = 0.51), suggesting that no major historical selective events had occurred. Similarly, none of the Ds for the individual large populations were significantly different from 0. The global F_ST for the large population samples was very large (0.444), with 44.41% of the variance occurring between populations and 55.59% within populations. A comparison of the fixation indices N_ST and G_ST revealed a strong relationship between phylogeny and geography, with N_ST being much larger than G_ST (0.743 and 0.427, respectively). The regressions of haplotype proportion (Fig. 3a) and haplotype diversity (Fig. 3b) with sampling latitude were both significant (R² = 0.475, P = 0.002; R² = 0.388, P = 0.006, respectively) giving additional support for the importance of phylogeographic structure in the species. When a single outlier was removed (corresponding to a northern site of secondary contact in New York, see Discussion), more of the variation could be accounted for by latitude alone (R² = 0.752, P < 0.001; R² = 0.483, P = 0.002; for haplotype proportion and haplotype diversity, respectively).

Phylogeographic analyses

All of the haplotypes recovered from N. serricornis were successfully connected in a single network at the 95% confidence level by tcs. The tcs analysis, however, placed haplotype ‘1’ as the most ancestral, based on its frequency in the overall data set and its range of geographic distribution, but as described above, the Bayesian phylogenetic analysis placed the root with a group of haplotypes in Pennsylvania and New York. A Bayesian analysis rooted with haplotype ‘1’ had a ln(likelihood) of 2370.51, much lower than the 2602.41 for the Bayesian tree rooted with C. cornutus. The haplotype network was therefore polarized for the NCA using the results from the more likely Bayesian tree rather than the tca analysis. The haplotype network was partitioned into 32 one-step clades (Fig. 4), 19 two-step clades, 9 three-step clades, 4 four-step clades, and 2 five-step clades (Fig. 5). Of these 66 clades, 13 exhibited a significant correlation between geography and phylogenetic structure in the NCA. The specific outcomes from the NCA inference key can be seen in Table 2.

Overlaying the un-nested haplotype network on the sampling geography yielded a close match between the two (Fig. 1). Six major clades were apparent: one covering Tennessee, Illinois, Indiana, Michigan, Wisconsin, and Ontario; a second centred in central Kentucky with haplotypes also occurring in southeastern Ohio and in Georgia; a third radiating out from North Carolina into Pennsylvania, southeastern Ohio, and eastern Kentucky; a fourth, only represented in Georgia, a fifth clade, containing the ancestral haplotypes, spanning across Pennsylvania and into New York; and the last clade occurring along the eastern seaboard, from North Carolina up to Connecticut and Massachusetts, and into Maine.

The results from the global amova are presented Table 3. All F_ST values were highly significant, with P values < 0.001 being returned by the arlequin permutation tests. Of most significance for historical interpretation, almost 20% of the variation occurs among clades.

Intra- vs. interspecific divergence in Nigronia

The mean sequence divergence within N. serricornis (0.02 substitution/site) is an order of magnitude smaller than it is between N. serricornis and Nigronia fasciatus (0.219 substitution/site), the only other species in the genus Nigronia. The degree of sequence divergence within N. serricornis and between the two species of Nigronia is consistent with COI sequence divergences from other orders of insects. Hebert et al. (2003a) found that 196 of 200 lepidopteran species they examined exhibited interspecific COI sequence divergences in excess of 3%, and in Coleoptera, the most closely related order for which COI sequence divergence has been examined, 93.2% of the congeneric species pairs (n = 891 pairs) displayed interspecific divergences in excess of 4% (Hebert et al. 2003b). Therefore, it is reasonable to assume that N. serricornis exists as a single species across its range, despite its limited dispersal ability (Heilveil 2004). The Bayesian analysis also supported the monophyly of N. serricornis with respect to N. fasciatus and did not recover any significant subdivisions within the species.

Hewitt (2000) states that ‘Rapid colonization by this leading edge model in any part of the world should produce areas with reduced genomic variability.’ As Hewitt predicted, haplotype diversity was reduced in recently deglaciated areas; however, latitude only explained 39–48% of the variation in haplotype diversity in our study. Other factors such as secondary contact clearly play an important role in shaping patterns of haplotype diversity, as almost twice as much variation could be explained by latitude when a single point of secondary contact was removed from the correlation.

The major clades of N. serricornis in eastern North America

According to the Bayesian analyses rooted with Corydalus cornutus, clade II-19 (containing haplotypes 53–59; Fig. 5), represents the most ancestral haplotypes found in N. serricornis. The populations from which these haplotypes were recovered occur in central Pennsylvania and New York, the former being south of the Wisconsinan glacial maximum (Fulton 1989), and therefore may have served as a refugium. Typically, geographically widespread high-frequency haplotypes are assumed to be ancestral (Crandall & Templeton 1993; Castelloe & Templeton 1994), implying that haplotype ‘1’ in clade I-1 (Fig. 5) was the most ancestral: it occurs in 90 of the 344 animals sequenced and covers approximately 1100 km of the range of N. serricornis. An examination of the geographic distribution of clade I-1 (Fig. 1, in yellow), however, reveals that the clade predominantly occurs in areas that were uninhabitable during much of the Wisconsinan glaciation and therefore evolved more recently. The ancestral-like characteristics of clade I-1 will be discussed below.

The largest nesting level of the haplotype network consists of two main clades, V-1 (‘Ancestral’) and V-2 (‘Derived’) (Fig. 5). The populations represented in the Ancestral clade (the most basal haplotypes) stretch from Pennsylvania, New York and Maryland across to central Kentucky and into Missouri, representing the initial colonization of the eastern USA by N. serricornis. The Appalachian Mountain range, which neatly separates the Ancestral clade into clades IV-4 and IV-3, is likely responsible for the ‘restricted gene flow’ inferred by the NCA.

Within clade IV-4, a combination of populations from clade III-7 (‘central Kentucky’) and clade III-8 (‘Missouri’), there is evidence for past fragmentation followed by range expansion. According to Templeton (2004), the past fragmentation is further supported by the ‘larger-than-average’ number of mutational steps between the Missouri haplotypes (which are at the limit of the 95% CI for being connected to the rest of the network) and the central Kentucky clade. A similar pattern of geographically disjunct haplotypes was observed in Ambystoma maculatum Shaw (Phillips 1994).

Clade IV-3, the oldest of the four-step clades, experienced a contiguous range expansion out of Pennsylvania and into Connecticut and Maryland, with continued gene flow between the populations. More recently, however, allopatric fragmentation has occurred between these regions (NCA results for clade III-6, Table 3), potentially due to the poor dispersal ability of the species.

The Derived clade represents a second major wave of colonization by N. serricornis. Based on the polarization of the haplotype network, haplotypes in this clade appear to have spread out from North Carolina northwards along the coast, southwestward to Georgia, and westward to Tennessee and into the northern Midwest. The populations in this clade have undergone contiguous range expansion, and it is probable that the western migration out of North Carolina skirted the southern edge of the Appalachians. The Derived clade breaks down into clades IV-1 and IV-2, both of which experienced restricted gene flow and isolation by distance. The former of these, clade IV-1 (‘North Carolina’), radiates out from North Carolina into Tennessee, Kentucky, Ohio, Pennsylvania and Maryland. Part of the populations in this clade reside in eastern Kentucky, across the Cumberland Mountains and the intervening lowlands, a probable cause of the restricted gene flow inferred for this clade.

Clade IV-2 on the other hand, covers a vast amount of geographic space from the East Coast south to Georgia and north into Michigan and Wisconsin. This clade displays some of the most interesting patterns observed in this study. Clade IV-2 is comprised of clades III-5 (‘Georgia’), III-4, and III-1. The Georgia clade is restricted in distribution to Georgia and displays no phylogeographic structure. In contrast, clades III-4 and III-1 show two strikingly different patterns. Within clade III-4, is a large group of haplotypes that are spread out along the East Coast (clade II-9, ‘Coastal’). The Coastal clade as a whole has experienced restricted gene flow, possibly influenced by changes in coastal habitat induced by the Wisconsinan retreat. As the Wisconsinan retreated, water was released into the ocean, moving the shoreline inland (Fulton 1989) and altering the connectivity between coastal populations of animals. Within the Coastal clade, clade I-15 has undergone contiguous range expansion. Gene flow within this clade has allowed for haplotype transfer between populations in Maryland, Connecticut, Massachusetts, New York, and Maine. Similar movement of haplotypes from the southeastern USA northward along the coast has been observed in A. maculatum (Zamudio & Savage 2003).

The phylogeographic patterns of clade II-1 (‘the Tennessee clade’) are driven by clade I-1. This clade experienced a past fragmentation coupled with a range expansion that began in Tennessee and pushed north into Illinois, Missouri, Indiana, Wisconsin, Michigan and Ontario (Fig. 1, pictured in yellow). Similar to the findings of Seddon et al. (2001) and Alexandrino et al. (2002), the NCA made no inferences about the rapidity at which the range expansion took place. When the haplotype network is overlain on the sampling geography, it becomes clear that the expansion occurred over a short geological time period. The paucity of unique haplotypes recovered from the northern populations in this clade supports a rapid rate of range expansion with little gene flow occurring between the newly established populations and those found further south.

The recent, rapid, range expansion of the Tennessee clade also explains why clade I-1 displayed basal-like characteristics, as previously mentioned. During a rapid range expansion with little to no gene flow between populations, newly colonized populations have a limited number of haplotypes from which to be founded, allowing a single haplotype to spread over a large geographic area. If the range expansion occurred recently, novel haplotypes that have evolved will not have dispersed far, and the number of populations sampled from within the boundaries of the range expansion will bias the haplotype frequency towards the single ‘colonizing’ haplotype. Additionally, few mutations will have had time to accrue in any single lineage. All of the haplotypes occurring in clade I-1 for N. serricornis are single-step mutations that, based on their geographic distribution, have arisen since the Wisconsinan glacial retreat. We are currently working on obtaining sequences for other genes to examine some migration patterns at a finer scale.

Only a single haplotype was recovered from either population in Wisconsin and it is of Tennessee origin. The lack of genetic diversity in Wisconsin populations, in spite of reasonable sample sizes (n = 16 and 18), suggests that N. serricornis did not use the Driftless Area as a glacial refugium during the Wisconsinan glaciation unlike chipmunks (Rowe et al. 2004). Rather, N. serricornis followed the south-to-north migration pattern seen for most other organisms examined (e.g. Edmands 2001; Brant & Orti 2003; Zamudio & Savage 2003).

As predicted, the life history traits of N. serricornis have also shaped the population structure in the species. Out of the 13 clades for which an association could be found between phylogeny and geography, 10 clades experienced either population fragmentation or restricted gene flow. This preponderance of fragmentation suggests that individuals migrate between populations only in extremely rare instances, providing further support for limited dispersal in the species.

Phylogeographic congruence with other species in eastern North America

The only other study of postglacial range expansion by eastern North American aquatic insects was performed by Ross & Ricker (1971) on the plecopteran genus Allocapnia. Using distributional information together with morphological variation, Ross and Ricker stated that ‘only three species of Allocapnia (granulata, ninipara, and rickeri) appear to have moved northward through [northern Illinois and Indiana].’ Ross and Ricker felt that the slow, sandy streams present in Illinois and Indiana served as a barrier to migration for clean-water (i.e. dissolved-oxygen sensitive) species, such as Allocapnia illinoisensis. Ross and Ricker postulated that A. illinoisensis colonized Wisconsin and Minnesota via New York and Ontario, based on a ‘slight’ morphological variation found in A. illinoisensis from two Illinois streams and distributional information. Insects were obtained by solicitation, so it is unclear how intensely streams in this area were sampled and whether a priori assumptions about distribution influenced collecting effort. Interestingly, Ross and Ricker suggested that A. rickeri, another clean-water species, migrated through northern Illinois and Indiana, even though the species exhibited an almost identical distribution to A. illinoisensis (Ross & Ricker 1971). The molecular data from the pollution-intolerant N. serricornis suggest that dissolved oxygen (resulting from low flow and high sedimentation) levels in Illinois and Indiana streams were not barriers to migration for this species. Instead, the Tennessee clade migrated through the area en route to Michigan and Wisconsin. The inferred pattern of migration from Michigan into Ontario is identical to that observed for the ‘interior clade’ of the spotted salamander (Zamudio & Savage 2003), further supporting migration through Illinois and Indiana.

By overlaying the haplotype network on the sampling geography, points of secondary contact between two population expansions in central and northeastern New York (Fig. 1) become apparent. In these populations, haplotypes from both clade II-19 and the Coastal clade are present. The re-integration of haplotypes from these two clades artificially inflates both the proportion of haplotypes and the haplotype diversity above that expected for a similarly located population for either clade alone (Fig. 3).

Masta et al. (2003) point out that the phylogeographic patterns seen in aquatic animals may not reflect those of terrestrial animals. The major reason for this is that the distribution of aquatic organisms is often tied to hydrologic connectivity, as shown for golden crayfish (Fetzner & Crandall 2003). A good example of this relationship can be seen in the distribution of the central Kentucky clade (Fig. 1, pictured in dark green). Haplotypes from this clade occur in central Kentucky, Georgia and southeastern Ohio. The clade is noticeably absent from eastern Kentucky in populations that lie on a straight path between the central Kentucky and southeastern Ohio sites. The NCA did not find any significant patterns in the distribution of the central Kentucky clade. One explanation for this visual pattern is that subsequent to the initial colonization of southeastern Ohio, the central Kentucky clade became locally extinct in the eastern portion of Kentucky (similar to events described in Masta et al. 2003). Taking the aquatic nature of N. serricornis into account, however, a more plausible explanation becomes apparent: that the expansion of the central Kentucky clade followed river channels. A comparison of the distribution of this clade to a map of preglacial drainages composed by Mayden (1988) supports the probability of a drainage-mediated range expansion (Fig. 6). The population of N. serricornis sampled in central Kentucky is located in the Green River drainage basin, which, prior to the Wisconsinan glaciation, was closer by river distance to the population in southeastern Ohio than to either of the populations in eastern Kentucky. Drainage systems may guide population expansions in aquatic animals complicating the relationship between haplotype diversity and latitude. One solution to this complication is to perform the NCA using river distances, as opposed to latitudinal distance (e.g. Fetzner & Crandall 2003). Fetzner & Crandall (2003) suggest that river distance is the appropriate distance measure to use for organisms with an aquatic dispersal stage. Although the use of river distance would allow a better explanation of events in the central Kentucky clade, overall, terrestrial distance is more appropriate for use with N. serricornis, as dispersal occurs primarily during its terrestrial adult stage, and work on a confamilial showed extremely limited larval movement (Hayashi & Nakane 1989), further supporting the use of terrestrial distance over river distance.