Mitochondrial DNA (mtDNA) has been the workhorse of research in phylogeography for almost two decades. However, concerns with basing evolutionary interpretations on mtDNA results alone have been voiced since the inception of such studies. Recently, some authors have suggested that the potential problems with mtDNA are so great that inferences about population structure and species limits are unwarranted unless corroborated by other evidence, usually in the form of nuclear gene data. Here we review the relative merits of mitochondrial and nuclear phylogeographical studies, using birds as an exemplar class of organisms. A review of population demographic and genetic theory indicates that mitochondrial and nuclear phylogeographical results ought to concur for both geographically unstructured populations and for populations that have long histories of isolation. However, a relatively common occurrence will be shallow, but geographically structured mtDNA trees—without nuclear gene corroboration—for populations with relatively shorter periods of isolation. This is expected because of the longer coalescence times of nuclear genes (approximately four times that of mtDNA); such cases do not contradict the mtDNA inference of recent isolation and evolutionary divergence. Rather, the nuclear markers are more lagging indicators of changes in population structure. A review of the recent literature on birds reveals the existence of relatively few cases in which nuclear markers contradict mitochondrial markers in a fashion not consistent with coalescent theory. Preliminary information from nuclear genes suggests that mtDNA patterns will prove to be robust indicators of patterns of population history and species limits. At equilibrium, mitochondrial loci are generally a more sensitive indicator of population structure than are nuclear loci, and mitochondrial estimates of FST-like statistics are generally expected to exceed nuclear ones. Hence, invoking behavioural or ecological explanations of such differences is not parsimonious. Nuclear genes will prove important for quantitative estimates of the depths of haplotype trees, rates of population growth and values of gene flow.
Our understanding of the evolutionary history of populations has been dramatically enhanced by the acquisition of molecular data that have revealed the distribution of genetic variation within and among populations. Surveys of allozyme variation were the method of choice in the period from the late 1960s to the mid-1980s (Avise 2000). However, well-documented limitations of allozymes resulted in their disuse when it became feasible to survey variation directly at the DNA level; after all, what could provide a better record of evolutionary history than the blueprint of heredity itself? Restriction enzyme digests of organelle DNA were used for a few years, but these soon were replaced by direct sequencing of mitochondrial, chloroplast, and nuclear DNA (nuDNA). The field of molecular evolutionary biology was taken by storm with the realization that mitochondrial DNA (mtDNA) could be easily obtained from animals, was rapidly evolving, and was potentially informative at a variety of taxonomic levels. Subsequently, thousands of published studies have reached conclusions about population history, patterns of gene flow, genetic structure, and species limits, on the basis of mtDNA sequence variation.
Mitochondrial DNA phylogeography
Mitochondrial DNA is present in most cells in high copy number and is relatively easy, rapid, and inexpensive to sequence. For example, if well-preserved or fresh tissue samples are available, then sequences of a kilobase or more can be produced for 100 or more individuals, perhaps distributed over 10 or more populations (limitations stem mainly from costs and time to gather samples). If a rapidly evolving locus is chosen, a researcher could observe a ratio of haplotypes to individuals of 50% or more, providing sufficient variation from which to draw inferences. Thus, if the range of a species is reasonably well sampled, an overview of its genetic structure at that locus can be obtained. In particular, a phylogenetic tree of mtDNA haplotypes, rooted with a closely related taxon, will reveal whether closely related haplotypes occur only locally or throughout the range, i.e. whether the genetic structure of the population is viscous or characterized by substantial gene flow. This then, was the essence of phylogeography (Avise et al. 1987) and it continues to be an important component. However, many recent mtDNA phylogeographical studies often include estimates of parameters important in evolutionary processes, such as levels of gene flow, patterns of natural selection, dates and rates of diversification, and population growth. Modern phylogeography then has both pattern and process components.
The mtDNA indictment
Almost from the beginning of phylogeographical surveys, potential concerns were voiced about the exclusive use of mtDNA (Avise 1994). For example, the rapid rate of mtDNA evolution leads to recurrent substitutions at single base positions (homoplasy or ‘saturation’), which can obscure the signal of deeper history. However, the biggest concern was that all the genes in the mitochondrial genome evolve as a single linkage unit. Thus, no matter how many mtDNA genes are sequenced, the analysis yields a single gene tree. Single gene trees might misrepresent the organismal phylogeny for several reasons. For example, mtDNA reflects matrilineal history; this might be a biased portrayal of the overall lineage history if there were sex-biased dispersal. Also, if multiple population divergences or speciation events were closely spaced in time, a single gene tree might be ‘incorrect’ by chance due to the random nature of lineage sorting during the coalescence process. Despite these long-acknowledged concerns, hundreds of studies have drawn evolutionary inferences from mtDNA alone.
Recent reviews (Ballard & Whitlock 2004; Edwards et al. 2005; Bazin et al. 2006) have renewed concerns over studies that are based solely on mtDNA. As one might expect, the recommended check on mitochondrial gene trees is to compare them with multiple, unlinked nuclear loci (Rubinoff & Holland 2005). Although hypothesis testing is a vital part of the scientific process, the call for corroboration of mtDNA results appears to be conflated with an ad hoc opinion that such results are not valid without confirmation. For example, Edwards et al. (2005: 6552) commented that ‘in our view, maternally inherited mtDNA can never capture enough of a species’ history to delimit species on its own’ and that ‘mtDNA should not have priority over nuclear genes in avian species delimitation.’ This indictment of the use of mtDNA data to detect geographical/taxonomic patterns affects hundreds of studies. Should it be taken as a verdict also? The case for the primacy of nuclear variation for studies of phylogeography is not so clear to us. In particular, it seems to us that criticisms of mtDNA confound estimates of the two components of phylogeographical inquiry, pattern and process. For the many legitimate reasons to estimate the pattern of population history, mtDNA is likely the marker of choice because of its rapid coalescence time (Box 1). In contrast, for robust estimates of demographic variables, multiple independent loci are needed to reduce the process error associated with coalescence; the algorithms for estimating such parameters ‘treat the gene tree as a nuisance parameter, rather than the object ...’ (Dolman & Moritz 2006). In fact, mtDNA estimates of population size (Ne), gene flow (mNe), population growth (e.g. mismatch distributions), and divergence times (times to most recent common ancestry or population sundering) are associated with large confidence intervals (Edwards & Beerli 2000), and the solution to that problem is to analyse multiple nuclear genes to reduce the coalescent process error. But things are not so straightforward for the problem of detecting patterns of recent diversification. Below we explore the use of nuclear loci to investigate phylogeographical history.
Nuclear gene approaches to phylogeography
Introns and single nucleotide polymorphisms (SNPs): phylogeography or population genetics?
In theory, intron sequences can be treated in a manner identical to mtDNA sequences: gene trees can be inferred and individual haplotypes used as the terminals in phylogenetic analyses. However, because nuclear mutation rates are generally lower than mitochondrial ones (perhaps by a factor of 10; Brown 1983), there will usually be less variation per sequenced base for nuclear than for mitochondrial sequences. On the other hand, neutral variation scales as µNe (where µ is mutation rate), and because Ne for nuclear loci will frequently be four times that of mitochondrial loci, the overall level of intron variation compared to mitochondrial may be approximately 40% or so. Thus, one might have to sequence two to three times as many nuclear bases as mitochondrial to uncover equivalent numbers of variable sites and, hence, to produce a tree with equivalent resolution (i.e. ratio of observed haplotypes to individuals). Of course, producing this gene tree requires that there be no recombination and that one can somehow ‘phase’ or ‘sort’ (determine the co-allelism of) the two sequences that occur in each individual — not necessarily a trivial undertaking. Additionally, the fourfold increased effective population size of nuclear genes over mitochondrial ones results in an important difference between introns and mtDNA, because it means that mtDNA will detect recent splits for which nuclear loci will not have had time to sort (see Box 2).
A study of a single nuclear locus will not be as productive as a comparable analysis of a mitochondrial tree would be because of the reduced amount of variation (e.g. haplotypes per individual) in nuclear sequences (and particularly in SNPs). Consequently, nuclear studies will inevitably focus on estimates of parameters taken across loci. For example, one might estimate the size (µNe) and growth rate of each sampled population and the amount of gene flow (mNe) between population pairs; this can be done efficiently with software currently available and the confidence intervals on the results will be greatly improved over those associated with mtDNA (or any single locus) variation. Basically, sampling additional individuals in an mtDNA study allows us to reduce the error associated with the shape of the individual coalescent tree, but it cannot let us estimate the error associated with the coalescent process at that single locus. Multiple nuclear loci do solve this problem, but the associated cost is that we have essentially eliminated the ‘phylo’ part of phylogeography and converted the study of geographical variation into one of retrospective population genetics. This is the great legacy of Kingman's (2000) coalescent: population genetics now has the mechanics to look back in time to help us capture process history. But, the well-resolved individual gene trees are gone and we are left with vectors of µNe's, matrices of bidirectional estimates of mNe and pairwise estimates of divergence times. And these are not current estimates; they are products of events occurring over a period four times longer than for mtDNA. This could be important for species living in areas, such as north temperate zones, in which several temporally distinct environmental changes might have left genetic imprints still discernable. Hence, we argue that a well-resolved mtDNA gene tree ought to be the starting point for a phylogeographical investigation.
Gene frequency approaches
At present, most nuclear studies of geographical variation involve gene frequency approaches such as microsatellites or amplified fragment length polymorphisms (AFLP). The salient advantage of augmenting an mtDNA study of phylogeography with microsatellite markers is the potential to obtain multiple gene estimates of evolutionary patterns and processes (replicates in a statistical sense). However, in practice, this is not trivial. A basic component of phylogeography, literally, is the superposition of a gene tree onto geography. In most phylogeographical studies of microsatellites, the ‘phylo’ component is usually seriously compromised. This is because it is difficult to infer an actual phylogenetic tree from microsatellite frequencies without making a number of heroic assumptions involving clustering individuals into populations for analysis, and rooting the resulting distance trees (Box 3).
It is possible to detect geographical structure with microsatellites using the individual as the unit of analysis; this can be done with programs such as structure (Pritchard et al. 2000). However, this method is based on the detection of departures from Hardy–Weinberg and linkage equilibria, i.e. detecting a Wahlund-type effect. Such structure, however, is not necessarily the signature of vicariance or long-term isolation of populations. In fact, any reduction in gene flow that allows the divergence of allelic or haplotypic frequencies will be found with this methodology, given adequate sample sizes (Proctor et al. 2005). Thus, for example, sampling small island or fragmented populations, or populations at the opposite ends of a long linear range, can yield significant results. The program structure may be quite useful for detecting units of population management (e.g. Latch et al. 2006), but it cannot be used alone to identify historical entities.
Classification of phylogeographical pattern
Avise et al. (1987) outlined several general outcomes of mtDNA phylogeographical studies. In one of these, the topology of the gene tree indicated deeply divided groups that are geographically allopatric (Category I). At the other extreme, the mtDNA haplotype tree was geographically unstructured, with haplotypes from disparate areas interspersed on a shallow tree. Other potential outcomes included geographically structured shallow trees, and deep haplotype trees that were not geographically structured.
A question of interest is whether one or more particular mtDNA phylogeographical outcomes are more likely to misrepresent the evolutionary history of the populations under study. A potential way to answer this question is to determine how often analyses of nuclear genes contradict mtDNA results. Table 1 summarizes the possible general outcomes of comparisons of mtDNA and nuclear gene trees. Interpretation of these outcomes requires discussion.
Table 1. Comparisons of mtDNA and nuclear gene trees in phylogeographic studies. Only major geographical inferences are treated here
Reciprocally monophyletic groups
No reciprocally monophyletic groups
possible lineage sorting error?
selective sweep? NUMT?
inappropriate clustering analysis or over-interpretation of structure (or similar) software?
nuclear coalescent time = 4× mitochondrial coalescent time? expectation of FST (mt) > FST (nu) due to differences in Ne?
female gene flow much less than male gene flow?
Reciprocally monophyletic groups
A1: congruent no conflict A2: incongruent conflict*
C: conflict (= false negative mtDNA† or false positive nuclear‡)
No reciprocally monophyletic groups
B: conflict (= false negative nuclear§ or false positive mtDNA¶)
D: no conflict
Geographically structured trees for both mtDNA and nuclear markers
Geographically structured mtDNA trees suggest long-term population isolation and offer new perspectives on species limits. In the majority of cases studied to date, geographically structured mtDNA gene trees make geographical and taxonomic sense. For example, analysis of mtDNA haplotypes of the fox sparrow (Passerella iliaca) recovered the same four parapatric groups as those initially defined by early naturalists as species based on plumage colouration (Zink 1994). Avise & Nelson (1989) found two clearly distinctive and parapatric groups of the seaside sparrow (Ammodramus maritimus), with each group representing multiple subspecies. These two groups are congruent with numerous other mtDNA patterns in this area, suggesting a common vicariant origin (Avise 2000; see Riddle et al. 2000 for a similar example). How could this congruence arise if mtDNA patterns were not correctly recovering evolutionary history? This is not to say that information from morphology or nuclear loci should not be considered. However, well-supported geographical patterns of mtDNA differentiation should be sufficient to provide a strong inference of evolutionary history.
Nevertheless, there are cases in which mtDNA and nuclear gene trees might be positively conflicting (e.g. outcome A2 in Table 1). To qualify as conflicting, mtDNA and nuclear trees should each have significant statistical support, and substantial geographical conflict; that is, a major portion of the geographical range identified by an mtDNA marker as part of clade ‘X’ must be identified as part of alternate clade ‘Y’ by the nuclear markers, perhaps due to chance errors of lineage sorting (e.g. Funk & Omland 2003). If a single nuclear and mtDNA locus conflict (not owing to lack of resolution in either), and there were no other information to resolve the conflict, one should consider the lineage history unresolved. If multiple nuclear loci suggest a congruent pattern that contradicts the mtDNA tree, one would conclude that the latter was an erroneous indicator of lineage history. Of course, there must be some ultimate reason for this conflict and that in itself is worthy of investigation. Lineage sorting, hybridization between lineages followed by asymmetrical gene flow, and even mitochondrial gene capture of one population are potential causative phenomena.
mtDNA structured, nuclear loci uninformative
Outcome B (Table 1), in which a shallow mtDNA gene tree shows reciprocal monophyly but nuclear loci do not support it, could result from several alternative phenomena. First, we note that in the case of transient nonequilibrium following fragmentation, this result is actually consistent with differences in coalescence time for nuclear and mitochondrial loci (Moore 1995; Palumbi et al. 2001; see Hudson & Turelli 2003). Therefore, if the mtDNA gene tree shows reciprocal monophyly and the phylogroups are closely related, one should not expect nuclear loci to recover the phylogeographical splits because of their longer coalescence times (e.g. ‘intermediate past’ in Box 2). This relationship was termed the ‘three-times rule’ by Palumbi et al. (2001), and suggests that if the coalescence ratio (distance of clade to sister clade divided by nucleotide diversity of clade) exceeds three, nuclear loci should recover lineage splits. This situation represents an inability of nuclear markers to detect a recent fragmentation event because they are overly lagging indicators (Box 1).
A second explanation is that, due to differences in the effective population sizes of mitochondrial and nuclear markers, equilibrial FST-like statistics are expected to be divergent when there is appreciable geographical population structure (Crochet 2000). In particular, the expectation of FST (mtDNA) is greater than the expectation of FST (nuDNA) for values of FST greater than zero and less than one (see Appendix).
A final explanation for a shallow mtDNA tree and no nuclear support is that there is extreme philopatry in females, but nuclear gene flow through males. Bowen & Karl (2007) referred to this discrepancy as ‘complex population structure’ in an analysis of the population structure of marine turtles. In some waterfowl species, males follow females from the wintering grounds to the female's natal area. This could result in mtDNA structure without nuclear support, the latter indicating male gene flow. Such a situation would represent a failure of the mtDNA marker to reflect accurately the overall population structure. However, this is a fairly unlikely occurrence; the female philopatry would have to be nearly 100% for a sufficiently long period of time for the mtDNA genetic structure to coalesce, on the order of 2Nef generations. For waterfowl with very large population sizes, this might predate the most recent glaciations. For most birds other than waterfowl, it is female dispersal distances that exceed those of males (e.g. Greenwood & Harvey 1982), not the reverse.
mtDNA and nuclear loci unstructured
Geographically unstructured mtDNA trees (outcome D in Table 1) are common in studies of temperate birds (Zink 1997). Here, several explanations are possible. First, it is possible that all populations exchange migrants, resulting in the geographically intermingled haplotype tree; that is, there may be ‘normal’ isolation by distance and no phylogeographical pattern for either nuclear or mitochondrial markers. In this case, however, for any weak geographical structuring of populations at all, most mitochondrial measures of the structure will exceed equivalent nuclear ones (Appendix). Second, it is possible that dispersal recently ceased between geographically isolated areas, and that lack of reciprocal monophyly results from shared ancestral polymorphisms at all markers (e.g. ‘recent past’ in Box 2). Many of the avian species that have been studied to date occur in the Holarctic, and probably represent populations with post-Ice-Age range expansions; an unstructured haplotype tree is one of the expected topologies for this type of demographic history. That is, both mitochondrial and nuclear markers are lagging indicators of the current demographic structure of the population (Box 1).
A commonly voiced concern about unstructured mtDNA haplotype trees is that they often occur in species with geographical variation in some phenotypic characteristics. For example, north temperate birds, with an average of 5.5 subspecies per species, exhibit only 1.8 reciprocally monophyletic mtDNA groups (Zink 2004). Thus, phylogeographical studies have frequently failed to ‘recover’ morphologically defined avian subspecies. Morphological variation could exist without mtDNA structure because morphological traits are polygenic, often under selection, and can evolve much faster than neutral mtDNA markers can coalesce and achieve reciprocal monophyly (Box 1). However, many avian subspecies reflect arbitrary, subjective divisions of character clines, rather than discrete evolutionary groups (Remsen 2005).
mtDNA unstructured, nuclear loci structured
Outcome C in Table 1 is potentially the most ‘damaging’ to the reputation of mtDNA studies. In this situation, nuclear loci recover geographical differentiation that is not found in the mtDNA gene tree. It is difficult to find valid examples of this outcome in the still developing literature of mtDNA vs. nuclear comparisons; one potential case is discussed below. In some cases, conflict has been reported due to inappropriate analysis; in the Alaskan song sparrow (Pruett & Winker 2005), mean population divergences, rather than individual haplotypes, were clustered.
It is the case that times to reciprocal monophyly for individual loci are stochastic, and it is possible, by chance, to sample an instance of a slowly coalescing mtDNA tree and to find a rapidly coalescing nuclear one. However, given the magnitude of the differences in Ne, it will be rare that a nuclear gene tree coalesces first, given reasonable sample sizes of individuals. Suggested additional causes, if it is the mtDNA that is misleading, are the inadvertent sequencing of a nuclear copy of an mtDNA gene region, or a selective sweep.
Nuclear copies of mitochondrial genomes are known to exist (NUMTs); however, an analysis of the completely sequenced chicken genome suggests they are not common and are mostly composed of small fragments (Pereira & Baker 2004). For a protein coding gene, old nuclear copies are easily detected by searching for stop codons, an equal distribution of substitutions across all three coding positions, and heteroplasmy. Only very recent transpositions of mtDNA to the nucleus should pass all these filters. For the noncoding control region, NUMTS are harder to detect, but if suspected, a coding gene for the same individuals could be examined with the above criteria to check for consistency with NUMTs. Note however that, in theory, an erroneously sequenced nuclear copy of a mitochondrial gene ought to yield a result consistent with the other nuclear genes in the study, not in conflict. Conflict due to NUMTs may be a spurious suggestion.
Selective sweeps (Ballard & Whitlock 2004) are expected to result when a novel advantageous haplotype arises that quickly replaces all other haplotypes at that locus in the population; this will remove all previous evidence of population structure and history and replace it by a single, closely related family of haplotypes. Selective sweeps might occur for both mitochondrial and nuclear markers and obscure geographical signal in either case; they are potentially more damaging for mtDNA because there is no reason to expect the phenomenon to co-occur across unlinked nuclear markers. Thus, in a multilocus nuclear study, sweeps should not affect multiple genes in a consistent pattern; consequently, in a nuclear study there are internal checks available, whereas there are none in a mitochondrial survey. Selective sweeps are frequently invoked but hard to reject in the absence of the very large sample sizes required for statistical tests of selection. Note, however, that sweeps cannot go where genes do not flow. That is, if there has been a strong barrier to gene flow for a sufficient period of time to result in reciprocally monophyletic clades, it is difficult to understand how the selectively advantageous haplotype can cross the barrier to sweep through the alternate population. To erase a prior mitochondrial signal, there must be some gene flow. Theoretical and empirical work suggests that weak natural selection against slightly deleterious mutants does not obscure phylogeographical signal (Neuhauser & Krone 1997; Zink 2005; Zink et al. 2006).
The avian data — phylogeographical patterns
A sample of avian phylogeographical studies involving mitochondrial and nuclear DNA was reviewed to assess the correspondence between markers in detecting phylogeographical patterns (Table 2). Most of the comparisons involved mtDNA sequences and microsatellites. These comparisons can be difficult to judge for several reasons. First, with enough microsatellite loci and individuals, even slight frequency differences become statistically significant (see also Box 3). This can give the impression that there is substantial population structure when in fact, one must judge whether an FST of 0.03 (for example) is biologically significant (e.g. Hedrick 2005). We were conservative in scoring results such that only discrete clustering of microsatellite data and mtDNA reciprocal monophyly counted in the ‘A’ category. However, we provide brief notes that indicate which marker provided greater differentiation in frequencies (if any). Second, it is possible for a single study to show multiple patterns, such as when there are two major groups (outcome A) with insignificant differences among populations within groups (outcome D); to avoid replication, we did not report the second pattern for such cases. Furthermore, some studies reported conflicting patterns. For example, Johnson et al. (2007) reported multiple comparisons among Gyrfalcon populations (Falco rusticollis), and in some mtDNA, differences were greater than microsatellites, and in others, the reverse pattern was observed. Cases such as these are relatively rare.
Table 2. Studies involving both mtDNA and nuclear loci in birds
In Fig. 1, we provide a simplified overview of the results of our tabulated (Table 2) mitochondrial vs. nuclear (mostly microsatellites) comparisons. We used the relevant FST-like statistic reported in the original study as an approximate indicator of the presence or absence of deep geographical structure. Unfortunately, many alternative FST-like statistics are available, and those reported for mtDNA and nuDNA for a particular species are not always equivalent. We took an FST value of 0.2 or greater (20% of genetic variance distributed among regions) as indicative of significant structuring (Category I of Avise 2000). Thus, points in region A of Fig. 1 are examples in which both mitochondrial and nuclear markers indicate substantial geographical structuring; points in region D are cases in which neither class of marker identified any deep structure. Regions B and C are areas of apparent conflict; however, points in region B are actually consistent for the cases of shorter coalescent time of mitochondrial markers following recent isolation (i.e. leading indicator) and of differing theoretical expectations of FST with equilibrial population structure (see Appendix). From this point of view, only points in region C are truly unexpected.
In general, there was good correspondence between mtDNA and microsatellite markers (Table 2), given the theoretical expectations we have outlined (Fig. 1). Many studies found that the two sets of markers were either concordant or the mitochondrial estimate of a statistic exceeded the equivalent nuclear one. One apparent exception to the general concordance is the study by Eggert et al. (2004) in which mtDNA control region sequences of Lanius ludovicianus did not detect a difference between mainland samples in southern California, whereas microsatellites did. However, mtDNA revealed slightly greater differentiation between the southern mainland and those on the Northern Channel Islands (California), which the authors suggested was due to sex-biased gene flow. A concern about the Eggert et al. (2004) conclusions involves the small number of mitochondrial control region bases studied (250), and the correspondingly low number of haplotypes resolved (six in 90 individuals); in addition, the sample sizes from two of the localities were small. It would be more informative if a well-resolved mtDNA gene tree were compared against microsatellite loci.
Several studies found shallow but structured mtDNA trees that were not supported by microsatellites (outcome B). In many cases, the authors explained the apparent conflict by assuming sex-biased gene flow. We feel the theoretical results outlined in this study represent a more parsimonious explanation. In general, few authors acknowledged the potential effect of differing coalescence times for mitochondrial and nuclear loci. That is, the perception that microsatellites evolve more rapidly than mtDNA (e.g. Newton 2003) may have led authors to assume that a lack of microsatellite structure was due to bias in gene flow, not to the greater lag time for nuclear loci. This view confuses the mutation rate with the coalescent time (e.g. Box 2). One study did note that accounting for these theoretical expectations brought disparate mitochondrial and microsatellite statistics into agreement in tawny owls (Strix aluco; Brito 2007).
Some studies reporting hybridization revealed the value of having nuclear information to complement mtDNA data (Vallender et al. 2007). Because mtDNA in animals is maternally inherited, having nuclear information greatly aids in interpreting the mtDNA results. Bensch et al. (2002) describe a situation in which mtDNA typing is unable to reveal the hybrid nature of some individuals that is apparent at microsatellite loci. This is an expected result.
Because of the many difficulties in interpreting microsatellites, studies involving direct sequencing are essential. This would make the data far more comparable and amenable to the same coalescence methods. For example, Jennings & Edwards (2005), using 30 anonymous nuclear loci, found that a majority supported the mtDNA tree for some Australian grass finches. Although many loci did not recover the mtDNA topology, this was not surprising because the mtDNA tree shows that the speciation events were closely spaced.
Of the few studies to date that have compared sequences from mtDNA and nuclear loci, that of Bensch et al. (2006) is the most provocative. These authors compared sequences from cytochrome b (cyt b) and four nuclear loci within and among some closely related species of Old World warblers (genus Phylloscopus). The mtDNA (cyt b) gene tree shows reciprocal monophyly at the species level, little structure in the haplotype tree for Phylloscopus trochilus and therefore low nucleotide diversity, and a deep phylogeographical division (5.4%) within the chiffchaff (P. collybita) between P. (c). brehmi and the other putative subspecies. Two important results were reported. First, they reported that two nuclear loci (AFLP-WW1, GenBank DQ174644 to DQ174677, AY139181, AY139180; CHD-Z, GenBank 174556 to 174581) showed relatively deep phylogenetic structure in willow warbler (Phylloscopus trochilus) that was not apparent in the mitochondrial gene tree. Second, they found that none of the nuclear genes recovered the deep mtDNA split within the chiffchaff. Therefore, these observations suggest significant conflict between mitochondrial and nuclear markers (outcomes C and B, respectively, in Table 1). In fact, Bensch et al. (2006: 170) concluded ‘... that the use of mtDNA alone can give a much too simplified and biased view of the phylogenetic history of a species group ...’.
We conducted independent analyses on the sequences. Because there was no variation at CHD-W, and because MC1R was not considered in conflict with mtDNA, these loci are not discussed further. It is important to note that few individuals were sequenced for the same genes (for some valid reasons), rendering exact comparisons difficult. The nuclear sequences were aligned with the software clustal_x (Thompson et al. 1997), whereas the cyt b sequences (GenBank Accession nos DQ174582 to DQ174608) are coding and already aligned; neighbour-joining trees were constructed from p-distances using mega4 (Tamura et al. 2007), and 1000 bootstrap replicates were used to generate a consensus tree.
We recovered the same basic tree structure (not shown) for WW1 as that shown in Bensch et al. (2006), showing three clades of willow warbler and one for the chiffchaff. This seems to suggest that there is significant geographical structure in willow warbler that was ‘missed’ by the cyt b analysis. More importantly, however, what is lacking from the published trees of Bensch et al. (2006) is information on the geographical locality of the specimens. Inspection of the geographical distribution of samples indicates that there is no geographical structure to the nuclear tree, exactly as the cyt b tree suggests. A possible explanation is that the authors sequenced three alternate copies of this anonymous nuclear locus. Thus, the only ‘conflict’ is in the apparent depth of the WW1 gene tree relative to that for cyt b. Analysis of Bensch et al.'s (2006) CHD-Z sequences shows a paraphyletic relationship between P. trochilus and P. collybita, as reported by Bensch et al. (2006). A paraphyletic relationship between the two species is exactly what would be expected for a locus that is lagging the mtDNA gene tree. Thus, the ‘conflict’ here is in the higher nucleotide diversity for the CHD-Z gene relative to cyt b. Lack of phylogeographical pattern is revealed by both mtDNA and nuclear loci. It is of interest to explore the apparent difference in diversity reported for mtDNA and nuclear data, but there is no evidence that it represents a clear case of outcome C (Table 1).
Bensch et al. (2006) interpreted a significant Tajima's D value to mean that the lower cyt b diversity is a result of purifying natural selection. It is known that this statistic also captures information on population expansion and many if not most Palearctic birds show such signatures (Drovetski et al. 2004; Zink et al. 2006). We subjected their cyt b data to McDonald–Kreitman (M–K, McDonald & Kreitman 1991) tests using dnasp (Rozas et al. 2003). Neither of the relevant tests were significant (P. trochilus vs. P. collybita, Fisher's exact test P = 0.21; P. trochilus vs. P. fuscatus, P = 0.46). In addition, we subjected the data to the R2 analysis for population growth (Ramos-Onsins & Rozas 2002) and found it highly significant. Therefore, it is most likely that these cyt b sequences were taken from a significantly expanding population, and that there is no evidence that natural selection has influenced the variation or differentiation of mtDNA sequences.
One of the principal goals of phylogeography has been to determine whether species consist of one or more independently evolving units, termed phylogroups, and to determine their interrelationships. If the objective is to determine the geographical or taxonomic limits of recently evolved groups, mtDNA is the molecule of choice because of its high variability and rapid coalescence time. It can be useful to sequence a nuclear locus as well, to ascertain whether the mtDNA tree is misleading, but our review suggests this will be rare. In cases in which a structured mtDNA gene tree is accompanied by unstructured nuclear trees, the former ought to be taken as indicative of lineage divergence; mtDNA is the leading indicator, nuDNA the trailing one. A few situations do exist in which the mtDNA gene tree will provide a misleading signal of recent phylogeographical history (Table 1), but nuclear loci will do so in many more instances. The more important question concerns how many of the existing mtDNA estimates of phylogeographical pattern are biased to the extent suggested by Edwards et al. (2005). Preliminary indications are that most mtDNA results can be taken as reliable indicators of geographical population structure, phylogeographical patterns, and species limits (Table 2, Fig. 1). However, investigation of other evolutionary questions, such as comparing depths of trees, estimating coalescence times, quantifying gene flow and population growth, and resolving phylogeny above the species level, will require multiple loci to provide robust estimates. Of course, the degree to which mtDNA estimates of these parameters will be erroneous has yet to be determined. Thus, modern evolutionary genetics studies that wish to describe both phylogeographic patterns and evolutionary processes will require multiple loci. We strongly urge that nuclear sequences, and not frequency-based approaches such as microsatellites, complement mtDNA data for those cases.
We thank A. Jones, A. Pavlova, J. Klicka, K. Barker, S. Lanyon, and B. Barber for comments on the manuscript. Four anonymous reviewers contributed constructive comments.
Box 1 Demographic and genetic population structures: leading and lagging indicatorsNatural populations are characterized by demographic and genetic structures; given long-term equilibrium, these correspond to each other. Following a demographic change, the genetic structure of a population only gradually reaches its new equilibrium; the time course of the approach is governed by the coalescent process. During this period of non-equilibrium, the signatures of events such as demographic fission, population bottlenecks, or growth can be found in the genetic variation in the population; the genetic structure contains information about the population's recent demographic history.?For vicariant events, such as Avise's (2000) category I phylogeographic pattern, haplotypes are shared between daughter populations immediately following the splitting. During the ensuing coalescent process, mutations that result in new haplotypes are restricted to one or the other of the daughter populations; these may increase in frequency and eventually the shared ancestral haplotypes become rare. If the two vicariant populations differ in size, then the smaller is expected to become monophyletic (coalesce) first, leaving the other population paraphyletic due to incomplete lineage sorting (a potential example is Spizella breweri and S. taverneri; Klicka et al. 1999). Eventually the two populations become reciprocally monophyletic.?For populations that are not in demographic/genetic equilibrium, different classes of markers carry information about divergent time scales and, in fact, there is a duality between the demographic and genetic structures of populations. Demographic data gathered today, such as dispersal observations and population growth rates, are a leading indicator of the future genetic structure of the population. Neutral genetic markers are a lagging indicator of prior demographic population structure. Because time to coalescence is a simple, positive function of effective population size, and given that Ne for nuclear loci is approximately four times Ne for mitochondrial loci, mitochondrial and nuclear markers are informative about demographic and dispersal structure in the intermediate (mtDNA) and distant (nuDNA) past. Traits under directional selection sweep to fixation much more rapidly than neutral lineages coalesce (i.e., independent of population size); hence they may identify relatively recent barriers to gene flow.
Box 2 Relative efficacy of molecular markers: mutation rates vs. coalescent timesFor neutral genetic markers sampled from a population of stable size, the sampled time scale (coalescent time) is determined by the effective size of the population and is independent of the mutation rate of the marker. Consequently, as a result of the approximate four-fold difference in Ne between nuclear and mitochondrial genes (Box 1), lineage sorting will be complete for mtDNA for some vicariant events for which it will not yet have occurred for nuclear markers. Thus, one expects to observe cases in which geographically allopatric or parapatric populations are reciprocally monophyletic for mtDNA sequences but paraphyletic or polyphyletic for nuclear DNA sequences (introns) or other markers (e.g. microsatellites). The most parsimonious explanation of such phenomena is that they are consistent with the intrinsic population genetics of the markers; that is, by themselves such observations do not require explanation in terms of divergent patterns of maternal versus paternal gene flow or hybridization. In such cases, it is not possible for nuclear data to ‘confirm’ mitochondrial data: mtDNA is a relatively ‘leading’ indicator and nuDNA a ‘lagging’ indicator.?The mutation rate of the sampled genetic marker determines the density of observed changes on the sampled coalescent tree and hence its resolution in terms of the observed number of haplotypes or alleles, but does not affect its depth in time. Thus, a mitochondrial coding region may have a lower mutation rate than a non-coding control region, and an intron may have a lower rate than a microsatellite, but, in both cases, this only affects the number of changes inferred on the tree, not its temporal scale. The mutation rate does affect the ease with which a tree can be rooted; too rapid a rate can obscure the signal needed to determine the position at which an outgroup attaches to the ingroup network. For example, for mitochondrial control region sequences, sequences from too distant an outgroup may be effectively randomized with respect to the ingroup. For microsatellites, rooting is often impossible because trees of relationship among alleles typically are not used, nor are outgroups routinely examined; this is due to an excess of homoplasy in microsatellite allelic states due to the slippage mutation mechanism.
Box 3 Microsatellites: issues of rooting and sampling An integral component of phylogeography involves the comparison of a rooted haplotype tree with the geographic distribution of samples. For DNA sequence data, mitochondrial or nuclear, the evolutionary sequence of base substitutions along homologous strands can be inferred to yield an unrooted network of haplotypes using phylogenetic methods. The addition of an outgroup to the study allows the network to become a directed gene tree of haplotypes that were obtained from individual organisms. For microsatellite alleles, which are assumed to differ by an integer number of repeats, there are usually many fewer microsatellite states at a locus than there are mtDNA haplotypes in a network. Moreover, there is considerable homoplasy (owing to convergence and parallelism to the same number of repeats) in these alleles and assigning an outgroup to a unique one of them requires the observation of limited overlap in states between the outgroup and ingroup and the assumption of a stepwise model of mutation. These are often dubious assumptions and consequently microsatellite studies are often unrooted; that is, there is no phylogeny of haplotypes to superimpose on a map. For example, Haig et al. (2001) derived an unrooted neighbor-joining (distance) tree from 17 RAPD loci that portrayed Mexican spotted owls (Strix occidentalis lucida) as the sister group to northern plus California spotted owls (S. o. caurina and S. o. occidentalis, respectively). However, Haig et al. (2004) subsequently reported on mtDNA sequences from these same birds plus an outgroup species, and found a gene tree in which the northern spotted owl was sister to the California plus Mexican spotted owl clade. That is, the mtDNA data set included a root and was preferred to an unrooted microsatellite result.?The second major problem with microsatellite data is that the unit of analysis frequently becomes a pooled sample of individuals (usually from a specified geographic area), not the individual haplotype. Branching diagrams that accompany many microsatellite phylogeographic studies generally are the result of clustering of the matrix of overall genetic similarities between population pairs. Because the population samples are pooled individual genotypes, any one haplotype might be genealogically closer to a haplotype in another population than to others in the same population. Thus, these clustering diagrams cannot represent an actual ‘tree’ showing phylogenetic relationship or pedigree because the clustered objects could be polyphyletic or paraphyletic groupings of genotypes, not monophyletic entities that have unique patterns of ancestry and descent.?
If the unit of analysis becomes the population and trees are obtained using clustering procedures, then in fact trees will be found that appear to indicate hierarchical arrangements among the populations, whether such hierarchy exists or not. For example, in a case of pure isolation-by-distance, the magnitude of genetic differentiation between pairs of population samples will be a function of the geographic distance between the samples. Thus, in the example shown here, there is isolation by distance for a character trait (e.g. allele, haplotype, or microsatellite frequencies). Given inevitable irregularities in the geographic sampling scheme, clustering will ‘identify’ hierarchical units determined by the size of sampling gaps, rather than by history, and the relative proximity of sampling locations will determine the hierarchical pattern recovered. For example, the presence or absence of a sample from Punta Escondida leads to either two (dashed and solid) or three (dashed, dotted, and solid) major clusters of localities in a UPGMA analysis. This, in turn, might lead an investigator to postulate erroneously the existence of two or three subspecies.
Barrowclough and Zink have been collaborating on studies of avian population genetics and phylogeography for over 30 years.
FST for nuclear and mitochondrial markers. If mitochondrial and nuclear estimates of FST (or similar variance-partitioning statistics) had identical expectations, empirical estimates ought to scatter around the diagonal in a plot such as the one shown in Fig. 1. If FST based on mitochondrial genes led those based on nuclear genes as an indicator of recent isolation, data points ought to fall below the diagonal; conversely, if nuclear markers were a leading indicator of isolation, vis-à-vis mtDNA, then points ought to fall above the diagonal. We can calculate the expected relationship between nuclear and mitochondrial estimates of FST, following the isolation of populations, for a simple genetic model. For such a case, FST(t) ª 1-e−t/2N; that is, the time course to equilibrium (FST = 1) scales as the negative exponential of the ratio of time to effective population size (Wright 1969; Nei 1975). Because the effective population size of mitochondrial genes is approximately one fourth that of nuclear genes (Birky et al. 1983), we have FST(nu) = 1-e0.25*ln[1–FST(mt)]. This is plotted in Fig. 1. It represents the path along which parametric values of mitochondrial and nuclear FST should proceed, starting from FST = 0 at time equal zero, to FST = 1 at time equal infinity.
Crochet (2000) pointed out that for the case of gene flow at equilibrium in a simple island model of population structure, estimates of FST based on mitochondrial loci ought to exceed estimates based on nuclear genes, assuming Ne(female) ª Ne(male). In particular, for gene flow rates, m, small compared to one, FST(nu) = FST(mt)/[4–3 FST(mt)]. This equation is also plotted in Fig. 1. It represents the set of equilibrial relationships between mitochondrial and nuclear parametric values of FST, given m << 1; each point on the curve corresponds to a single value of mNe. Note that this curve always falls on (at FST equal zero and one) or below the diagonal; thus, mitochondrial markers are a more sensitive indicator of population structure than nuclear ones.