Field sampling

The sampling took place at a total of 28 sites (Fig 1). A majority of these sites were located within the current known distribution of Lyme disease occurrence in Southern Québec, [27, 29, 32, 33]. The other sites were located in areas were Borrelia has not been detected but within the geographical range of P. leucopus in Québec (Fig 1). A total of 96 individuals of white-footed mice and 53 individuals of deer mice were captured in forest patches during the summer of 2011 using Sherman live traps (Table 1), of which respectively 87 and 43 specimens were successfully sequenced for all five mtDNA markers. All specimens were brought back to the laboratory for tissue sampling and genetic analyses.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Location of the 26 study sites sampled during the summer 2011 in southern Québec.

An additional site (Saguenay) is located further north-east and is not shown on this map.

https://doi.org/10.1371/journal.pone.0144112.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Geographical coordinates of the sampling sites, and sample size for P. leucopus (P. l.) and P. maniculatus (P. m.).

https://doi.org/10.1371/journal.pone.0144112.t001

Ethics Statement

All procedures were approved by the Québec Government (SEG permit #2011-05-15-014-00-S-F), the McGill University Animal Care Committee (AUP#5420) and the Comité de déontologie de l’expérimentation sur les animaux de l’Université de Montréal (#11–054).

Species identification and sequencing

For each specimen, we extracted DNA from tail tissues using the standard Quick-Lysis protocol [34], and all specimens were identified to the species using species-specific primers in multiplex polymerase chain reaction (PCR) [25, 35]. We then amplified five mtDNA regions using the primers described in Table 2. The amplifications were carried out in a 50 μl volume comprising 5 μl of 10x reaction buffer, 4 μl of dNTP solution at 2.5 mM, 20pM of each primers, 1.5 u of Taq DNA polymerase and 50 to 300 ng of DNA. The DNA amplification protocol was 94°C for 1 min, 35 x (94°C for 45 sec, 54°C for 1 min 30 sec, 72°C for 45 sec) and 72°C for 7 min. The PCR products were then sequenced at the McGill University Genome Québec Innovation Centre. We visualised the chromatograms with the program 4Peaks 1.7 [36] and aligned the sequences with the program ClustalX 2.1 [37]. We removed ambiguous sections at the beginning and the end of each sequence, as well as the gaps and repetitions susceptible to be polymerase reaction errors. Sequence alignments for all markers combined are available as supplementary material for P. leucopus (S1 Table) and P. maniculatus (S2 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Primers used for amplifying the five mtDNA regions used in this study.

https://doi.org/10.1371/journal.pone.0144112.t002

Data analysis

We first estimated numbers of haplotypes, private haplotypes, parsimony-informative sites, and segregating sites, as well as nucleotide diversity and haplotype diversity with DnaSP version 5.10.1 [38] and the R package ‘PopGenome’ version 2.1.0 [39]. For phylogeographic analysis, the optimal substitution model was then determined with the program ModelTest [40]. We reconstructed phylogeographic trees for each species and all five mtDNA markers separately with the R package ‘ape’ version 3.1–4 [41] using the neighbor-joining (NJ) algorithm [42] applied to pairwise genetic distances among individuals (function nj) and corrected with the K2P model [43] for unequal base frequencies [44]. In addition, PhyML [45] was also applied to the five mtDNA markers, based on the same substitution model [43], and using the NJ tree as a starting tree. For better visualization, the trees were artificially rooted on the longest branch (midpoint rooting [46]) with the R package ‘phangorn’ version 1.1–1 [47], and a majority-rule consensus was computed (function consensus) to identify the supported clades. The robustness of separate phylogeographical trees was assessed with bootstrap values [48] computed from 100 replicates, and tree resolution was measured by counting the ratio of non-zero length branches over the total number of branches. We performed an incongruence length difference (ILD) test [49] with PAUP* 4.0 [50], and test of congruence among distance matrix (CADM) [51, 52] with the R package ‘ape’ to compare the markers with one another and determine whether they led to congruent phylogeographic patterns (function CADM.global). We then ran a posteriori comparisons using Mantel tests [53] to test for the significance of differences between pairs of markers (function CADM.post).

We built haplotype networks with the program Haplotype Viewer [54]. Whereas a phylogeographic tree links individuals, a haplotype network links genetic sequences, thus merging all unique haplotypes at a single node. For both species, we performed a clanistic analysis to evaluate the clans and slices purity [55], where clans are groups obtained by cutting a single branch (bipartition), and slices are groups obtained by cutting two branches (tripartition) in an unrooted network. We calculated the clan and slice purity by counting the number of haplotypes from each sampling site in a group divided by the size of this group (a value of one indicating that a given clan—or slice—contained only haplotypes from a single locality). We also calculated the clan and slice diameter as the maximum distance (number of mutations) between two haplotypes from the same locality in a group; this value was standardized by the total diameter of the network for comparison purposes. We used a Student t-test on all clanistic variables to test for significant differences in the haplotype networks of the two Peromyscus species.

To further detect differences in the geographical pattern of genetic variability between the two species, we computed population networks among localities (nodes in the graph) by considering only shared haplotypes [56]. Namely, two populations (sites) are connected by an edge in the network only when they have at least one haplotype in common. To account for unequal sampling sizes, we estimated the connection probability with a resampling approach in which a single haplotype per site was randomly selected at each iteration prior to constructing the network. This procedure was repeated 1000 times to derive pairwise connection probabilities among sites, for each species and each marker. We used two indices to compare these population networks: the graph density (the number of connections in the graph relative to the number of possible connections), and the node degree distributions (where the degree of a node is the number of edges connected to that node). These two indices were compared between the two Peromyscus species with Student t-tests and non-parametric Mann Whitney tests, respectively.

We determined possible demographic changes in population size or selective regimes for both species independently as well as for both shores of the St. Lawrence River with three different neutrality test and computed mismatch distributions [57] to detect population expansion. To test for deviations from neutrality, we calculated Tajima’s D [58], Fu’s F_S [59], and Ramos-Onsins and Rozas’ R₂ [60] statistics separately for the five markers, as well as for all markers combined. We assessed statistical significance of each test with 1000 replicates using either Arlequin version 3.5.2.1[61], DnaSP version 5.10.1 [38], or the R package ‘Pegas’ version 0.7 [62]. The mismatch distributions were computed using the function MMD with the R package ‘Pegas’ [62]. We also computed Fst-values among all pairs of populations and between shores for each species with all markers combined, and assessed statistical significance with 1000 permutations in Arlequin version 3.5.2.1 [61].

Finally, we estimated divergence times for the split separating the north and south shores of the St. Lawrence River for the two species using a Bayesian analysis and a relaxed-clock dating implemented in BEAST version 1.8.2 [63]. To do so, the 87 D-Loop sequences of Peromyscus leucopus included in the present study were combined with all sequences listed in Shipp-Pennock et al. [19], and the 43 Cyt B sequences of Peromyscus maniculatus were combined with those listed in Dragoo et al. [20] to determine phylogeographic clades corresponding to past vicariance events. In each case, the combined dataset was trimmed to compare sequences of equal lengths. To calibrate the molecular clock for each mtDNA marker, the corresponding sequences of Peromyscus leucopus (respectively Peromyscus maniculatus) were used as outgroup, and the split between both species was estimated at 3.5 Mya [64]. The analyses were run for 10,000,000 generations, sampling model parameters every 1000 generations and consensus trees were produced with TreeAnnotator v. 1.8.2 after elimination of 20% of the trees as burn-in. The resulting trees for both species were visualized in FigTree 1.4.2.

Congruent phylogeographic patterns

It has been shown that mtDNA markers may be biased, for instance through differential male/female dispersion rate. The use of several mtDNA markers improves the detection of significant phylogeographic signals, and they can display patterns congruent with those found using nuclear markers [65, 66]. Here, we found that the phylogeographic structure was congruent among the five mtDNA markers we used. For both Peromyscus species, samples from the south shore of the St. Lawrence River and samples from the north shore fell in distinct groups, and results of Fst tests between both shores were significant, which confirms our hypothesis that the St. Lawrence River constitutes a major geographical barrier to dispersal for Peromyscus individuals. Besides this strong separation, we found no further clear spatial structure in our data. Individuals from a given site were seldom grouped together, especially in P. leucopus.

Different mtDNA sequences may also exhibit varying degrees of ancestral polymorphism and thus not reflect consistent coalescent histories [66], which supports the need for considering more than a single marker in phylogeographic studies. Here, we found that the variability of the COIII marker was similar to that found in another phylogeographic study on P. leucopus and P. maniculatus in Québec [67]. The variability level observed in P. eremicus in Baja California for the Cyt B and COIII was about ten times larger than for P. leucopus and P. maniculatus in Québec, likely because the former study included samples from a much larger geographical area [22]. Likewise, Shipp-Pennock et al. [19] obtained sequences twice as variable as ours for P. leucopus using the D-Loop in a study that included specimens sampled throughout the eastern United States and Canada. With respect to the numbers of haplotypes, other studies have identified approximately ten times more than we did using the D-Loop in P. leucopus across the United-Sates [18]. However, we found proportionally more haplotypes than Dragoo et al. [20], based on Cyt B sequences of P. maniculatus. These results illustrate the benefits of using multiple markers at a smaller geographical scale, but also show that different mtDNA regions may behave differently in different species.

Range expansion and demographic changes

We detected the signature of a recent expansion of white-footed mouse populations that displayed a significantly distinct genetic pattern than the one observed in the deer mouse, which has long been established in Québec. Haplotype network analyses clearly showed that P. maniculatus was more structured across the sampling area than P. leucopus. Namely, haplotypes from the same site tended to be more clustered with respect to localities in P. maniculatus than in P. leucopus. Most haplotypes defined by the combination of all markers were unique in P. maniculatus, while P. leucopus exhibited a large number of shared haplotypes between sites. Shared haplotypes are generally considered as ancestral retention and provide evidence for continuous migration between populations [10, 68].

Using population network analyses we also found that P. leucopus had widespread haplotypes without any barrier other than the St. Lawrence River, whereas P. maniculatus only shared a few haplotypes between sites. These results are consistent with a recent northern expansion of white-footed mice detected in Michigan [9] and in Southern Québec [16]. Similarly, by comparing the phylogeographic patterns of twelve sympatric species of frogs, the genetic homogeneity observed throughout the sampling area was attributed to a recent range expansion during favorable climatic conditions [66].

The low genetic diversity we observed in P. leucopus in Southern Québec is consistent with a recent colonisation and a rapid range expansion [69]. On the other hand, the high diversity of closely related haplotypes with no clear geographic structure and the lack of apparent gene flow in P. maniculatus confirm that these populations have established in the region for a longer time, an observation corroborated by other field studies [9]. Populations of P. maniculatus appeared to be more structured and more diversified than populations of P. leucopus sampled in the study area. This clear phylogeographic structure suggests that P. maniculatus reproduces mostly within established populations, and that gene flow between these populations is reduced compared with P. leucoupus. Interestingly, neutrality tests revealed a significant departure from neutrality in populations of P. maniculatus sampled from the south shore of the St. Lawrence River. The same populations of P. maniculatus also exhibited a mismatch distribution with a unique mode. This was not observed on the north shore, and not for P. leucopus, however. Significant deviations from neutrality can be explained by selection or demographic fluctuations such as population expansion following a bottleneck [59]. Likewise, populations that have been experiencing a recent demographic expansion [57, 70] or range expansion [71, 72] generate unimodal mismatch distributions. In the present case, we postulate that interspecific competition between the two Peromyscus species [73, 74] could result in a selective sweep of deer mouse populations on the south shore of the St. Lawrence River, following the recent northward range expansion of the white-footed mouse [9, 16]. These results further reinforce the difference in their phylogeographic structures, especially if P. maniculatus were extirpated from territory newly colonized by P. leucopus.

Overall, our results are consistent with the hypothesis of a range expansion in P. leucopus, and have important implications for the spread of Lyme disease in southern Québec, as the risk of occurrence of B. burgdorferi was strongly associated with the presence of P. leucopus in our study area [27, 29].

Distinct lineages and implications for the emergence of Lyme disease

All five markers used in this study supported an ancestral divergence of mouse populations from the north shore of the St. Lawrence River and populations from the south shore, in both P. leucopus and P. maniculatus. When the sequences of the two species sampled in southern Québec were combined with those obtained in previous studies across North America [19, 20], we found that mice from both shores of the St. Lawrence River belonged to different ancestral clades, in the two species. Mice from the north shore of the St. Lawrence River originated from a Midwestern clade, while mice from the South shore originated from the East Coast clade. Previous studies have also shown that these clades represent different glacial refugia that have since then been expanding northwards independently [18]. Based on our results, the split between these clades corresponding to distinct shores of the St. Lawrence River occurred well before the Late Glacial Maximum. We estimated the time of this divergence at around 2.1–2.7 Mya, a value that must be taken with caution since we did not consider a third Peromyscus species as an outgroup for this analysis.

The St. Lawrence River thus appears to be a geographic barrier sufficiently strong to maintain the ancestral divergence between the two distinct lineages of P. leucopus in Québec, and likely distinct associated strains of B. burgdorferi. In line with this hypothesis, a phylogeographic pattern similar to that observed for the mice was detected in B. burgdorferi, with two main lineages of Borrelia (Midwest and North-East) that diverged since the Late Glacial Maximum across North America [75, 76]. These lineages possess different forms of the OspC protein, an antigen against which mammals can develop immunity [77]. Experimental studies have shown that P. leucopus immunity to B. burgdorferi is strain specific [78]. Since distinct lineages of B. burgdorferi geographically match distinct clades of P. leucopus in the United States, the two different clades of P. leucopus found in Québec are likely carrying distinct strains of the bacteria. We expect two strains of B. burgdorferi, that diverged since the Late Glacial Maximum, to prevail in Québec and that these strains will keep on spreading with the steady range expansion of P. leucopus in the province. It has been shown that B. burgdorferi subspecies manifest different clinical symptoms and responses to diagnosis in humans [79]. The existence of different virulence levels associated with different strains of B. burgdorferi and distinct clades of P. leucopus has significant implications for the monitoring of the emergence of Lyme disease in Québec, and the design of efficient strategies to lower the impact of this zoonosis on human populations.