Introduction

DNA polymorphisms in non-coding regions are widely used for phylogenetic inferences of species relationships. In addition, some non-coding regions also exhibit enough variability for intraspecific studies, that is, to analyze phylogenetic relationships of subspecies, varieties, and domesticated forms or to analyze the structure of populations. A relatively old field of research using these polymorphisms is the classical population genetic approach (Wright 1951) that is still widely used in modern studies (Bachmann 2001). The main focus of this review, however, will be on the use of non-coding DNA in phylogeography. Phylogeography tries to infer population histories in space and time from the extant geographical distribution of genetic polymorphisms. During the last few years, phylogeography has been an important area of research (Avise 2000), and since its foundation, the numbers of papers on this topic are steadily increasing, as can be seen in many recent issues of the major systematics and ecology journals.

As most non-coding parts of the plant genomes are free to vary without much restriction from selection (but see Halligan et al. 2003; Clark et al. 2006; Guo et al. 2007; Kelchner and Graham 2008) they can contain an ample amount of polymorphic sites (Bosch and Quandt 2008; Koch et al. 2008; Rein et al. 2008). However, this advantage also creates a major difficulty access these regions. Analyses today rely mainly on an initial PCR amplification step that requires prior knowledge of the flanking sequence regions where primers can be designed to bind. Therefore, the majority of non-coding regions used today in systematics and population biology are flanked by relatively conserved DNA regions (mostly genes), allowing the design of (nearly) universal PCR primers (White et al. 1990; Taberlet et al. 1991; Desmesure et al. 1995; Blattner 1999; Weising and Gardner 1999). Variable regions without conserved parts in a distance easily coverable by PCR are hard to access, as specific PCR primers have to be designed for each species or closely related species group under study. If no prior sequence information for a targeted region exists, this is tedious and often far beyond the technical capabilities of researchers working in organismic biology. However, current efforts to obtain sequences of the whole chloroplast genome for many plant groups will allow easier primer design for future studies, as chloroplast regions conserved in specific families or genera then can be simply found via nucleotide database searches. Moreover, non-coding DNA might contain small conserved regions involved in DNA transcription regulation (Clark et al. 2006), which can be used to bridge large parts of otherwise non-conserved DNA stretches via PCR or during sequencing.

Population genetics

Population genetics mainly studies contemporary gene flow and population structure. The basis for this field of research was laid by the work of Fisher (1930), Wright (1931, 1951), and Haldane (1932). Many factors influence the genetic structure of populations, as, for example, seed and pollen dispersal, breeding system, population size, genetic drift, selective pressure, and adaptations to habitats. The mating system is very important in this respect, since self-pollination leads to low gene flow between populations and, in the course of time, to a reduced number of polymorphisms in populations (Glémin et al. 2006), while outcrossing reduces genetic differentiation in and among populations. Furthermore, historical factors also, such as past genetic bottlenecks, contribute importantly to population structure, but are hard to access with these population genetic methods.

There are studies using a variety of genetic markers and dealing with a wide range of topics. After showing which markers are suitable for intraspecific studies, we will mention a few exemplary studies dealing with different topics associated with population or conservation genetics.

Genomes and markers

Generally, all DNA regions variable enough can be used for population genetic studies or intraspecific studies in general. Frequently, the organellar genomes are used, because chloroplasts and mitochondria are mostly uniparentally inherited in seed plants and thus have some great advantages over biparentally inherited nuclear markers. The main advantage is that there is typically only one allele per cell and organism, and consequently no recombination between two alleles occurs. Due to different dispersal distances, biparentally, maternally, and paternally inherited genomes also exhibit strong differences in genetic differentiation between populations. Especially, maternally inherited markers mostly show a much higher population subdivision (Petit et al. 2005).

The mitochondrial genome of plants is considerably larger than that of animals. Additionally, pronounced differences in size and organization of mitochondrial genomes exist among plant taxa. Intramolecular recombination, leading to complex genome rearrangements and, therefore, variable gene order even within single individuals, as well as duplications and deletions, are common (reviewed in Palmer 1992). Furthermore, base substitution rates in plant mitochondria are rather low (Wolfe et al. 1987), resulting in only minute differences within specific loci among individuals or even species.

Chloroplast genomes on the other hand exhibit a much more stable structure than those of mitochondria, and also higher substitution rates can be observed (Wolfe et al. 1987). However, chloroplast DNA (cpDNA) was for a long time considered too conserved for intraspecific studies (Banks and Birky 1985). This perspective changed at the beginning of the 1990s (Soltis et al. 1992), after a number of studies had found intraspecific or even intrapopulational chloroplast variation (Wagner et al. 1987; Milligan 1991). This intraspecific variation was shown to be high enough for population studies regarding gene flow (reviewed by McCauley 1995). An interesting approach is the contrast of paternally or biparentally inherited markers with maternally inherited markers. Using this combination, the ratio and the distances of pollen- vs. seed-based gene flow can be measured (Dong and Wagner 1994; Latta et al. 1998; Fénart et al. 2007).

In gymnosperms, the situation is somewhat different. Here, chloroplasts are inherited mainly paternally and are, therefore, dispersed through pollen and seed, while mitochondria are mainly maternally inherited and thus dispersed via seeds only (Wagner 1992). Since pollen is normally distributed over far longer distances than seeds (Liepelt et al. 2002), mitochondrial markers exhibit a much stronger population differentiation than chloroplast markers and are important characters used for population genetic studies in gymnosperms (Johansen and Latta 2003), sometimes also used in connection with cpDNA markers (Chiang et al. 2006).

Chloroplast DNA regions that are often used in infraspecific studies encompass various intergenic spacers and introns, for which universal primers exist. A comparison of the variability at some non-coding cpDNA regions was reviewed by Shaw et al. (2005, 2007). Among the most frequently used cpDNA regions are the trnL intron and the trnL–trnF intergenic spacer, also often used in combination (Koch et al. 2006). Other variable chloroplast regions are the atpB–rbcL intergenic spacer (Hung et al. 2005; Chiang et al. 2006; Bänfer et al. 2006), and trnH–psbA (Xu et al. 2000).

Mitochondrial regions used in intraspecific studies of plants, especially gymnosperms, are, for example, the introns of the NADH dehydrogenase gene nad1 (Johansen and Latta 2003; Jaramillo-Correa et al. 2004; Chiang et al. 2006), nad7 intron 1 (Godbout et al. 2005), nad5 intron 4 (Liepelt et al. 2002) and the internal transcribed spacer (ITS) of mitochondrial ribosomal DNA (Huang et al. 2001).

Besides the organellar markers mentioned above, microsatellites (Tautz 1989; Tautz et al. 1986) or simple sequence repeats (SSR) are often used in population biology, and sometimes also for phylogeographic studies. Microsatellites are short tandem repeats of single bases (mononucleotide repeats; i.e., mostly runs of 8–20 T/A) or small DNA motives of up to five bases (in plants often GA/TC, GAA/TTC). Length variation at these loci originates mostly from slippage of the DNA strands during DNA replication, thus resulting in an increase or decrease of the number of repeat motifs. Uninterrupted long runs of SSR motifs (>10 repeats) normally show more length variation than shorter microsatellites or loci where a different base occur within the SSR motif, as the possibility of slippage increases with the number of consecutive repeated units. Microsatellites are much less common in plants than in animals (Lagercrantz et al. 1993). Nevertheless, they exist in the nuclear as well as in the organellar genomes. Microsatellites generally show high variability that can be useful in population genetic studies, when other sequences or fingerprint methods do not exhibit enough mutations (Tautz 1989; Powell et al. 1995). However, their high variability is also the major disadvantage of these markers, as microsatellite loci often show high levels of homoplasy, especially when distant populations or higher taxonomic levels are studied (Provan et al. 2001; Jakob et al. 2007).

Microsatellite primers designed for one species might as well amplify in closely related species (Guicking et al. 2006). However, for nuclear loci, no universal SSR primers exist, and even for the chloroplast genome universally variable loci are rare. Weising and Gardner (1999) published PCR primers for chloroplast SSR loci, which amplify in many plant families but do not always provide products with length variation. As repeat lengths might be quite different among different taxonomic groups, a locus selected for high repeat number in one species might be absent or much shorter and, therefore, less variable in most other taxa. This bias in selection procedure explains the often low variation at microsatellite loci when transferred to other taxa. Conversely, when using known sequences of taxa closely related to the target species to find SSR loci (e.g. using the wheat or rice chloroplast genome sequences to localize potentially variable loci in other grasses) it is often worth including loci with only small repetitive motifs in the screening process, as these might have evolved into longer and, therefore, more variable SSRs in the taxon under study.

Complete absence of variation at some chloroplast microsatellite loci was found, for example, by Provan et al. (1999) and Rendell and Ennos (2002). In such cases, it might be useful to re-sequence the locus to see if the absence of sequence length variation is due to shrinking or even complete loss of the microsatellite motif or if it is related to a population genetic bottleneck, erasing alleles of a potentially variable locus from the plant populations (Jakob et al. 2007). The latter should, however, influence variation at most loci in the genome, while loss of SSR motifs might be restricted to a single locus. Re-sequencing of different SSR alleles can also contribute to the understanding of the nature of length variation. As, not only the repeat number might vary, but also the motif itself, and insertions or deletions outside the microsatellite motif can change the length of the amplified fragment (Jakob et al. 2007), it seems advisable to generally check SSR loci, when comparisons among different species or even subspecies are conducted. A determination of the DNA sequence also seems necessary when two SSR loci occur in close proximity within an amplified locus. In this case, compensatory mutations, that is, the expansion of one repeat stretch is counterbalanced by shrinking of the second, can result in identical fragment length of different alleles. These complex SSRs should either be excluded from the analysis, or sequence identity has to be verified by re-sequencing, restriction digests, or single nucleotide sequencing (SNS) analysis (Guicking et al. 2008).

Frequently, fingerprint methods, such as amplified fragment length polymorphism (AFLP) (Vos et al. 1995) or random amplified polymorphic DNA (RAPD) (Welsh and McCleland 1990; Williams et al. 1990), were also used for population studies (see Nybom 2004 for a comparison between different marker systems). However, AFLP, RAPD, and similar methods are anonymous markers, and even if the majority of polymorphic sites may be within non-coding regions, the exact location is unknown. Therefore, we will not go into detail about these markers.

Analysis methods

One of the main inferences made in population genetics is how genetic variation is distributed within and among natural populations of interbreeding organisms to study gene flow, genetic drift, mating systems, mutation rates, and natural selection (Templeton et al. 1995). This is often quantified by F statistics, using fixation indices like F ST (Wright 1951) or related measures like G ST (Nei 1973). From FST values, gene flow between populations can also be calculated (Beaumont 2005). As these are classical methods, they are included in many population genetic computer programs (Pearse and Crandall 2004). The main drawback of many traditional population genetic analysis methods is that the models used for the calculations of population genetic parameters are based on a variety of assumptions. Some of them, like constant population sizes and random mating in a Wright–Fisher population, are rarely met in natural populations (Whitlock and McCauley 1999; Hey and Machado 2003). Templeton et al. (1995) refer to a further major limitation of the use of F ST, as the data used to estimate the F statistic often do not indicate which model of gene flow (e.g. “island model”, “stepping stone model”, “isolation by distance model”) is appropriate for the populations being studied, particularly as the different models are not necessarily alternatives (see also Lynch and Crease 1990; Hudson et al. 1992). Besides, the geographical genetic variation measured by F ST may not be caused by the current amount of gene flow but instead be shaped by events far back in time (Templeton et al. 1995).

An important analysis method is the hierarchical analysis of molecular variance (AMOVA) developed by Excoffier et al. (1992). AMOVA analyzes population subdivision through F-statistics by measuring the correlation between genetic variation drawn at different levels of a hierarchically subdivided population. These correlations can be influenced by several evolutionary forces, like mutation rates or migration (Excoffier and Heckel 2006). However, the procedure of AMOVA is based on several assumptions, like random mating and the absence of inbreeding, of which natural populations mostly depart. Moreover, amova needs an a priori definition of the hierarchical structuring of populations, which might rely on wrong assumptions of the investigator. This problem might be solved by new analysis programs like STRUCTURE (Pritchard et al. 2000) or BAPS (Corander et al. 2003) which are based on Bayesian analysis algorithms and either deduce population structure directly from the data instead of using predefined settings or allow at least the testing of different assumptions.

Another frequently used statistical analysis is Mantel’s test (Mantel 1967), which estimates the correlation between two distance matrices, for example, the genetic distance of the analyzed samples and the geographic distance of the collection sites.

For a review of population genetic methods and a compilation of software suitable for population genetic analyses see, for example, Pearse and Crandall (2004) and Excoffier and Heckel (2006). Additionally, Waples and Gaggiotti (2006) presented a review of methods to identify the number of gene pools or populations in a given sample.

Examples of studies

There are a number of questions that can be solved with the information about genetic structure in plants. It can be used to evaluate, for example, migration rates between populations, inbreeding coefficients, effective population sizes, spatial genetic structure, or hybridizations between taxa.

One fairly large field of research connected to population genetics is conservation biology. To preserve endangered species, information about their population genetic structure is important, because it can give hints about which populations are especially important to preserve (Maudet et al. 2002), or which measures should be taken to preserve a special population (Ellstrand and Elam 1993). The genetic structure of geographically very restricted species has been studied in many cases to evaluate possible genetic depletion. Genetic consequences of habitat degradation or fragmentation have also been studied (Kettle et al. 2007). For restoration of habitats and species reintroduction at a certain site, it is also often advisable to study the population genetics of the species, to choose the best source populations for seed transfer according to genetic diversity and possible local adaptations (Ramey et al. 2000; Smulders et al. 2000).

Other important fields of population genetics deal with the characteristics of invasive species and their provenance (Williams et al. 2005; Jahodová et al. 2007; Londo and Schaal 2007; Okada et al. 2007) or the study of recently originated allopolyploids (Abbott et al. 2007). Also, the population genetics of clonally growing species can be analyzed. For example, the relative importance of clonal versus sexual dispersal was studied by analyses of the population structure of sea grass (Alberto et al. 2005). Methods for assessing clonality were reviewed by Arnaud-Haond et al. (2007).

Population genetic data can also be combined with geographical and ecological data (for a review about landscape genetics and related methods, see Manel et al. 2003) or be used to infer mating systems and factors such as inbreeding depression (Michalski and Durka 2007).

Phylogeography

Phylogeography as a distinct discipline arose in the late 1980s and combines microevolutionary (population genetics) and macroevolutionary (phylogenetics, systematics) concepts with the distribution of genetic variation in space and time (Avise et al. 1987; Avise 2000). In principle, the same genetic markers can be used in population genetics and phylogeography. The goals and analysis methods differ, however, even if population genetic methods are sometimes also used in phylogeographic analyses. The major difference between both fields is that population genetics interprets differences in allele distribution under the assumption of recent gene flow, while phylogeographic analyses explicitly seek to find out the historical processes that shaped the extant distribution of genetic variation. Phylogeography can, therefore, complement the analysis of fossil remains, such as pollen, or give insights into the history of species for which fossil remains are scarce or indistinguishable from other taxa.

Phylogeography and its basic assumptions

To reconstruct a species’ history with a phylogeographic approach, the genetic variation within this species is organized into a genealogy and overlaid by the geographical distribution of the alleles of the marker region under study (Avise 1989). Genealogy here refers to the progenitor–derivative relationships among these alleles, mostly depicted via genealogical networks. The analysis then interprets patterns of congruence or incongruence between the extant geographic distribution of alleles and their genealogical relationships on the background of different recent and historical processes influencing the structuring of genetic diversity within and among populations, that is, geographic barriers, dispersal events, population size changes, and gene flow.

The basic assumptions of phylogeography are mostly derived from coalescent theory, the formal mathematical and statistical treatment of gene genealogies within and among related species (Felsenstein 1971; Griffiths 1980; Tavaré 1984; Hudson 1990, 1998). Coalescent theory describes the merging of allele lineages in common progenitor alleles when going back in time. It allows the recognition of the polarity (old vs. young haplotypes) from the topology of an unrooted tree or network together with frequencies of haplotypes. Castelloe and Templeton (1994) showed that tip haplotypes or clades (which are connected to the remaining network by only one connecting branch) are almost always younger than interior ones (which possess more than one connecting branch to the remaining cladogram). Thus, contrasting tips versus interiors strongly tends to contrast younger versus older alleles. Predictions from coalescent theory show further that older alleles should prevail in populations, and be characterized by a higher number of descending lineages and a geographically wider distribution than younger alleles (Neigel et al. 1991; Neigel and Avise 1993; Castelloe and Templeton 1994; Posada and Crandall 2001). Although often true, the validity of these assumptions, which were tested in simulation models (Castelloe and Templeton 1994), depends strongly on sampling design, as well as life history traits and differences in the history of the taxa under study (Jakob and Blattner 2006; Jakob et al. 2007). As alleles become extinct at random over time, unless the population permanently grows, the oldest alleles are not necessarily the most frequent ones within a taxon (Avise 2000). Also the number of descending alleles has to be considered critically, since a certain allele may not have any descendants (or not any more) within the species under study, but descendants might occur in close relatives, when incomplete lineage sorting, that is, the persistence of ancestral polymorphisms through speciation events, is present (Jakob and Blattner 2006). This can invert the time axis of a network and result in completely different conclusions about historic processes (Jakob and Blattner 2006; Liston et al. 2007).

The differences in genetic diversity among populations in different areas are used for the reconstruction of a species’ history. The rationale behind this is the observation that during an expansion of the distribution area of a species, genetic diversity declines with distance from the starting point or center of distribution. This matches the hypothesis that the number of different haplotypes per area should be reduced during population expansion through repeated genetic bottlenecks occurring on the leading edge due to low population sizes and repeated founder events (Hewitt 2000) and the lower penetration rates of newly arising or arriving alleles in areas already occupied by conspecific individuals. A greater evenness in geographic haplotype frequency distribution can be expected if expansion started from multiple sources (Song et al. 2006) or if the population expansion is quite old, resulting in the slow admixture of many haplotypes within specific areas. As demography is critically important for allele numbers and their distribution, different historical events can be reconstructed from the allele patterns. For example, the foundation of a new population after long-distance dispersal should start with a single allele (or a very low number of alleles), which might be quite common in the source population. Thus, this allele will be shared between both populations. All alleles evolving after the dispersal event will, however, be exclusive to the respective population and are private alleles or haplotypes of these groups, as long as no introgression occurs. Besides, the allelic richness should be much higher in the source population in comparison to the population that originated with the initial genetic bottleneck of a founder event. This situation of haplotype distribution can similarly be found in speciation events beginning with a strong bottleneck for the new species. The picture looks different if population differentiation or speciation happened through vicariance, that is, if large populations became separated. In this case we could expect nearly equal haplotype diversity in the sister taxa, and for a long time the occurrence of shared alleles (Hudson and Coyne 2002; Jakob and Blattner 2006; Syring et al. 2007). Also, hybridization might be recognized by haplotype distribution. If tip haplotypes are found in species where none of their progenitor alleles occur and/or if haplotypes are area-specific instead of taxon-specific, this is a strong indication for hybridization (Bänfer et al. 2006). For internal haplotypes, however, it is often impossible to discern incomplete lineage sorting, that is, shared alleles, from hybridization events. Another mechanism contributing to geographical haplotype differences is genetic drift, which might result in fixation of different alleles in fragmented populations. Such populations might show a low gene diversity, whereas diversity among populations is high (Jakob et al. 2007).

Genomes and markers useful for phylogeographic studies

Here, we will focus mostly on non-coding parts of the organelle genomes, as they are most frequently used in phylogeographic studies in plants. However, some recent studies also used the variation at coding or non-coding nuclear regions to infer intraspecific differentiation (Olsen and Schaal 1999; Morrell et al. 2003; Caicedo and Schaal 2004; Schmuths et al. 2004; Bartish et al. 2006; Joly et al. 2006; Koch et al. 2006; Schmid et al. 2006; Gurushidze et al. 2007). Nuclear loci might create problems in data interpretation as recombination occurs and, therefore, mosaic sequences might be included in a data set. These can distort gene or loci genealogies if they remain undetected. The recognition (and probably exclusion) of recombinant alleles is, therefore, crucial for data analysis (Schaal and Olsen 2000).

Uniparentally inherited organelle markers have specific qualities for phylogeographic studies, as effective population size should be reduced in these markers compared to nuclear markers, since they can be considered as effectively haploid (Birky et al. 1989; Petit et al. 2005). Smaller effective populations sizes should result in faster turnover rates for newly evolving genotypes, resulting in a clearer picture of past migration history than nuclear markers (Rendell and Ennos 2002; Hudson and Coyne 2002; Kadereit et al. 2005). Moreover, problems associated with recombination are mostly absent in these markers. Initially, mainly animal species were studied phylogeographically using mitochondrial markers (Avise 2000). These studies provided, for example, interesting insights into the origin and evolutionary history of modern human populations (Richards et al. 1998). Contrary to the situation in animals, the use of mitochondrial markers in plants, especially in angiosperms, is more restricted (Tomaru et al. 1998). Today cpDNA markers are the most commonly used markers in phylogeographic studies of angiosperms, although mitochondrial markers are widespread in studies of gymnosperm taxa (see above).

As no chloroplast region variable in all angiosperms or let alone all land plants exists, often several loci have to be tested to find regions with a suitable variability in the group under study (Borsch and Quandt 2008). These tests involve sequencing of different chloroplast intergenic spacers or introns for a certain number of individuals. Candidate regions are accessible via PCR using universal primers (Taberlet et al. 1991; Desmesure et al. 1995; Weising and Gardner 1999; variability of some regions reviewed in Shaw et al. 2005, 2007) or by screening published chloroplast genome sequences of closely related species, if available (Shaw et al. 2007; Jakob et al. 2007; Sacks and Louie 2008). As described earlier for microsatellite loci, it might be worth to look for core motifs with slightly repetitive structures, as these might have evolved into highly variable loci in the taxa under study.

Laboratory methods

Early studies of chloroplast variation often applied restriction fragment length polymorphism (RFLP) with fragment detection via Southern blotting (Wagner et al. 1987). PCR technique and development of universal chloroplast primers amplifying introns and intergenic spacers by Tablerlet et al. (1991), Demesure et al. (1995), and others, led to a growing number of studies finding more and more variation in the chloroplast genome. PCR–RFLP, that is, the PCR amplification of defined DNA regions and their digestion with restriction fragments, was thus often used (Tremblay and Schoen 1999; Stehlik et al. 2002). As the number of available restriction enzymes increased and PCR provided nearly unlimited amounts of DNA of the target loci, these studies could arrive at high numbers of detected polymorphisms and, therefore, the phylogeographic resolution could be nearly as good as with DNA sequencing (Stehlik et al. 2002). This method, although currently still cheaper than sequencing, involves a lot of hands-on time and is, therefore, successively replaced by direct sequencing of PCR products of the respective target regions. This leads to another increase in detected polymorphisms and genetic resolution. However, also with sequencers using classical Sanger sequencing technology (i.e., four gel lanes to sequence one DNA strand) like, for example, radioactively labeled sequencing reactions or detection on ALF and machines, single nucleotide sequencing (Guicking et al. 2008) allows high throughput without losing much information within the analyzed region.

Microsatellite loci are sometimes also used in phylogeographic studies. However, due to their high variability, microsatellites are prone to homoplasy (Doyle et al. 1998; Ingvarsson et al. 2003), which particularly complicates the estimation of gene genealogies. According to some authors, they should rather be omitted if a reasonable number of other mutations are present because of the high risk of scoring non-homologous characters (Provan et al. 2001; Jakob and Blattner 2006). This depends, however, on the genetic distances among the studied populations or taxa, as SSR homoplasy should increase with the age of divergence and accordingly, the genetic distance. Bänfer et al. (2006) and Jakob et al. (2007) used a novel two-step approach to combine length variation at chloroplast SSR loci with sequence variation of intergenic spacer regions of the chloroplast genome, by first building a backbone genealogy on the basis of the sequence-based chloroplast haplotypes and adding the SSR variation for each so-defined haplotype creating subhaplotypes. Assuming lower mutation rates in the non-repetitive parts of the genome, this approach allows the use of chloroplast SSRs without introducing much homoplasy into the data sets.

Anonymous markers like AFLPs or RAPDs are less frequently used in phylogeographic studies than sequence-based marker regions because of their unknown genomic background. However, they are useful in cases were other markers reveal only very low genetic variation and have been used to analyze, for example, population histories in Arctic and alpine plant species (Gabrielsen et al. 1997; Tollefsrud et al. 1998; Friesen and Blattner 2000; Zhang et al. 2001; Stehlik 2002; Kropf et al. 2006).

Estimation of haplotype relationships

Beside the geographical distribution of the haplotypes, their relatedness is critically important for phylogeographic studies. Therefore, phylogenetic trees are often used to display these relationships. However, intraspecific gene evolution cannot be accurately depicted by bifurcating trees, but has to be represented by gene genealogies, that is, multifurcating networks (Clement et al. 2000; Posada and Crandall 2001), as ancestral alleles mostly coexist with multiple younger descendants. This phenomenon frequently results in zero-length branches within bifurcating trees, indicating that tips and internal nodes of a tree are occupied by extant taxa, alleles or individuals. In theory this should not be the case in phylogenies on higher taxonomic levels due to the longer time since population separation and, thus, the loss of ancestral alleles. However, a quick look through currently published phylogentic analyses of closely related species complexes immediately shows that zero-length branches are common even in interspecific studies. Also in these cases, network approaches often provide more information than phylogenetic trees and should, therefore, be used instead of or in addition to tree-based analyses (Gurushidze et al. 2007). Posada and Crandall (2001) discussed these problems in their detailed review and referred to distinct network approaches and currently available software packages to calculate gene genealogies.

The coexistence of progenitor and derivative alleles in a tree also causes problems for programs used to estimate absolute or relative ages of taxa and nodes in a tree like, for example, R8S (Sanderson 2002). The relative age of haplotypes can, however, be assessed using predictions of coalescent theory and their genealogical relationships (see above). Each haplotype network provides a time axis, from the oldest haplotypes at central and internal positions towards the tips of the network, where the youngest haplotypes are placed. Thus, rooting of a network is not always essential, as, contrary to phylogenetic trees, the reading direction is already inherent in a genealogy. As haplotypes will ultimately get lost after some time within a certain species due to lineage sorting, they appear in a genealogical network as missing intermediates, that is, their prior existence can be inferred from the character state differences of neighboring alleles. If the analyzed individuals cover the variation within a species quite well, the existence and distribution of these missing intermediate alleles in a genealogical network allows the inference of population history, as during population expansion, lineage sorting and, therefore, the loss of alleles is low, while with shrinking population size or an extreme genetic bottleneck lineage sorting and, therefore, allele extinction steeply increases (Jakob and Blattner 2006). Comparable to branch length distributions in phylogenetic trees (Barraclough and Vogler 2000; Barraclough and Nee 2001), missing intermediates reflect parameters of the population history in gene genealogies and networks. Deviations from the predictions of coalescent theory can, therefore, be expected if taxa are analyzed, where some ancient alleles are still present in some individuals, or if taxa or populations are compared that experienced rather different evolutionary influences during their histories (Jakob and Blattner 2006).

As haplotype relationships in a genealogical network are calculated by a statistical parsimony approach that represents each mutation as a step in the network, it is necessary to have single mutations represented as single characters in the data matrix. In DNA sequence alignments of non-coding DNA, often insertions or deletions (indels) are the main informative characters. As these can partly be quite long but anyway evolved via a single mutation event, they have, therefore, to be represented as single alignment positions. This means that longer indels occurring in some individuals have to be shortened to a single character state in the alignment, which might cause problems when several informative positions occur in that stretch of DNA where present. In this case, it is sometimes necessary to code this part so that all information can be represented and adjust the necessary mutational steps in the network afterwards manually to represent the correct haplotype relationships. As already mentioned, it could be necessary to exclude parts of the data matrix from the analysis if homology of the alignment positions (or more general, characters) could not be estimated safely (Morrison 2008). This holds particularly true for microsatellite loci but also some other repetitive parts might cause problems. Homoplasy in genealogical networks results in closed loops, that is, relationships are represented by several different connections between the haplotypes. Although these loops can sometimes be resolved based on coalescence assumptions, this problem is not different from homoplasy in phylogenetic analysis (Kelchner and Graham 2008) and can best be solved by using marker regions with adequate variability, exclusion of non-alignable sequence parts or involving a kind of weighting scheme, as described before for the two-step procedure of network construction when microsatellites are analyzed together with non-repetitive DNA parts (Bänfer et al. 2006).

Phylogeographic analysis methods and their problems

As mentioned before, traditional population genetic approaches also, that deal with the spatial frequencies of alleles and are mostly based on equilibrium expectations derived from the theoretical model of population structure under neutrality theory (reviewed in Felsenstein 1982; Slatkin 1985; Slatkin and Barton 1989; Neigel 1997), are partly used in phylogeographic analyses. Particularly, the hierarchical analysis of molecular variance (AMOVA) by Excoffier et al. (1992) and Mantel’s test (Mantel 1967) must be named here (see above). To describe the genetic diversity of populations and their differentiation, several diversity measures can also be evaluated, for example, Nei’s gene diversity H (Nei 1987), the haplotype richness R (El Mousadik and Petit 1996), or the number and distribution of population-specific haplotypes, so called private alleles (Stehlik 2002).

Templeton et al. (NCA; Templeton et al. 1995; Templeton 1998 and references therein) developed a method specifically for phylogeographic purposes, called nested clade analysis that tries to include the principles and basic coalescence assumptions mentioned before. This method is supposed to be able to distinguish between recurrent gene flow and a variety of historical processes, like past fragmentation, long distance colonization events, and range expansions. There is also a computer program available (GeoDis; Posada et al. 2000), which, together with an inference key, implements NCA. Due to its simple usage and ready-to-use results, the method was, and is, still the most popular in phylogeographic studies. Recently, different approaches have been made to automate the whole procedure of NCA (Zhang et al. 2006; Panchal 2007).

Nested clade analysis is, however, not undisputed. First, it was proposed that the choice of a randomization strategy for NCA greatly affects the outcome of the analyses (Petit and Grivet 2002). Secondly, Knowles and Maddison (2002) stressed that, although statistical significance is computed for association between the geographical distribution of haplotypes and their nested clade, no confidence limits can be assessed for the interpretations drawn from the inference key. In an analysis of simulated data sets with NCA, they found that NCA, in most cases, did not identify the processes that were used to simulate the data. Templeton (2004) counter argued that NCA performs reasonably well concerning hypotheses of range expansions and past fragmentations. However, Panchal and Beaumont (2007) conducted another study on a larger number of real and simulated datasets by using the software provided by Panchal (2007). They found that NCA returned a high number of false positives (but see Templeton 2008). Another potential problem concerns possible local population extinctions. Hung et al. (2005) showed that this can lead to erroneous identification of long distance dispersal in NCA. For these reasons, Petit (2008) recommended to refrain from using NCA until the method is further evaluated. A probable solution might be the use of statistical phylogeography (Knowles 2004; Knowles and Carstens 2007) where a wide range of demographical and biogeographical processes can be accommodated, provided biologically realistic models are available. However, as currently no easy usable software package or at least a protocol for these methods exists, we assume that it will take some time until statistical phylogeography will become a widespread analysis method.

As mentioned above, it is sometimes found that haplotypes are not restricted to the species under study. One possible explanation for such shared haplotypes is hybridization, which was inferred, for example, in European species of Quercus (Petit et al. 2002), Betula (Palme et al. 2004), and Arabis (Dobeš et al. 2004). However, evidence is accumulating that incomplete or differential lineage sorting also contributes to the distribution of haplotypes among different species (Mason-Gamer et al. 1999; Wendel and Doyle 1999; Comes and Abbott 2001; Linder and Rieseberg 2004; Syring et al. 2007). This means that the population involved in speciation was of considerable size, or a new species was formed more than once, so that more than one haplotype of the progenitor species is also present in the new species. This pattern is common in the New World species of the genus Hordeum (Jakob and Blattner 2006), where virtually all interior haplotypes are shared among up to six species. We assume that the same phenomena will be found in many young or rapidly radiating species groups, when sampling is expanded and more close relatives are also included in phylogeographic studies. This is supported by findings that chloroplast alleles might survive for quite a long time. Jakob and Blattner (2006) estimated survival times of chloroplast alleles within the genus Hordeum to up to 4 million years (Ma). This value is an order of magnitude higher than in, for example, Mediterranean Senecio (Asteraceae) where minimum survival times of shared chloroplast haplotypes of 0.44 Ma were found (Comes and Abbott 2001). Therefore, it is not surprising that some of the polymorphisms can be much older than the species in which they are found (Hudson and Coyne 2002). Generally, haplotypes shared between different species lead to considerable problems for phylogeographic studies, as it is normally difficult or even impossible to discern between hybridization and incomplete lineage sorting (Bänfer et al. 2006). The history of each haplotype (e.g. geographical restriction due to a population bottleneck, expansion or even extinction) depends strongly on the fate of the species in which the haplotype occurs. For example, geographical patterns arisen during the history of an ancestral taxon may still be visible and may conceal historic processes, which underlay the distribution of genetic diversity of derived extant taxa. Thus, also for intraspecific analyses, it might be advisable to include a certain amount of closely related taxa and to check carefully the geographic distribution of shared haplotypes on different spatial scales, which may allow distinguishing between both processes (Jakob and Blattner 2006).

Examples of phylogeographic studies

The rationale of most phylogeographic studies is, for example, to identify putative Pleistocene refugia, re-colonized areas, migration and dispersal corridors and, thus, to explain the observed population structure and geographic distribution of genetic diversity as well as to enlighten the evolutionary history of a certain species. However, phylogeographic methods have also been applied to study speciation and hybridization events (Joly et al. 2006; Jakob and Blattner 2006) or the history of polyploid formation (van Dijk and Bakx-Schotman 1997).

Phylogeographic studies have been conducted in many European taxa, studying the number and location of potential ice age refugia and re-colonization routes (Johansen and Latta 2003; Rendell and Ennos 2002; Bartish et al. 2006), sometimes in conjunction with fossil pollen data (Petit et al. 2002). In North America, a number of studies have been conducted on the same topic (Soltis et al. 1997 and references therein; Johansen and Latta 2003; Griffin and Barrett 2004; Godbout et al. 2005). There are also studies on plant species with special distribution patterns, like arctic plants (Tremblay and Schoen 1999; Abbott et al. 2000; Abbott and Brochmann 2003; Alsos et al. 2005) or alpine plants (Holderegger et al. 2002; Comes and Kadereit 2002; Stehlik 2003; Schönswetter et al. 2005; Dixon et al. 2007), and the post-ice age re-colonization of the continents by trees in Europe and North America (Dumolin-Lapegue et al. 1997; Petit et al. 2002; Grivet et al. 2006). The number of comparative phylogeographical studies increased remarkably during the last years. More recently, these studies started to look for congruence among phylogeographical patterns of a number of different species sharing the same distribution range (Soltis et al. 1997, 2006; Taberlet et al. 1998; Calsbeek et al. 2003; Kadereit et al. 2005), or to analyze the phylogeographic pattern in widely distributed species, where the genetic consequences of the same event (e.g. the Pleistocene) could have been quite different for each of the subpopulations (Ehrlich et al. 2007). For reviews on the genetic consequences of the ice ages for plants from different regions of the world see Hewitt (2000, 2004).

All in all, after more than a decade of phylogeographic studies on European or North American taxa, the knowledge of the response of plants and animals to Pleistocene climate changes, of the localization of ice age refugia, and patterns of postglacial re-colonization of formerly glaciated areas, increased considerably. However, not much is known about the impact of Pleistocene climate oscillations on plant species in the mountain ranges of Asia and Africa. Generally, phylogeographic studies in Africa, Asia, and the southern hemisphere are still scarce compared to Europe or North America. Cannon and Mannos (2003), for example, described the current geographical patterns of genetic diversity and inferred the historical population dynamics of the stone oaks (Lithocarpus) in Southeast Asia. Hung et al. (2005) used a phylogeographic study to enlighten the change of genetic diversity within Lithocarpus konishii after an earthquake. Aizawa et al. (2007) analyzed the phylogeography of Picea jezoensis by a combination of mitochondrial and chloroplast markers. Takayama et al. (2006) scrutinized the phylogeography and genetic structure of Hibiscus tiliaceus, a pantropical plant with an interesting modus of seed dispersal. Nettel and Dodd (2007) used a phylogeographic study to reveal dispersal patterns of mangrove species along tropical coasts. Some phylogeographic studies on species from the southern hemisphere have also been conducted. For example, Shepherd et al. (2007) studied the volcanic and glacial impacts on the distribution of the forest fern Asplenium hookerianum and Gardner et al. (2004) studied the late Quaternary phylogeography of Metrosideros (Myrtaceae) in New Zealand. Most of the existing phylogeographic studies from South America concern tree species and/or species from the Andes (Pastorino and Gallo 2002; Marchelli and Gallo 2004, 2006; Muellner et al. 2005), while rainforest species were rarely included (Olsen and Schaal 1999), and studies of the taxa of the huge pampa and steppe areas extending between the southern Andes and the Atlantic coast are completely absent up to now. Generally, studies of tropical plant groups, no matter if herbs, trees or epiphytes, are comparably rare, owing mostly to the problem of getting the good population representation necessary for phylogeographic studies. Thus, in groups where it is hard to obtain even samples of single individuals, as mostly included in phylogenetic studies, phylogeographic data will, for a long time, not become available unless major efforts are put into collecting good population samples. We assume, therefore, that future phylogeographic analyses in remote or hardly accessible regions will mostly be restricted to economically important species (Olsen and Schaal 1999). However, we think that more studies conducted on species outside Europe and North America are necessary to elucidate species and vegetation history, particularly in the background of the extant global climate change.