Drawbacks With Use Of Microsatellites In Phylogeography Biology Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

Microsatellites are commonly used molecular markers in phylogeography, and many view them as superior to mitochondrial DNA (mtDNA) gene trees. Being based on frequencies of alleles, and not gene trees, microsatellites exhibit the same analytical drawbacks that resulted in the abandonment of allozymes in genetic studies of population history. I illustrate some these familiar drawbacks by reanalyzing microsatellite data on the song sparrow. Subspecies were previously evaluated with hierarchical analyses of molecular variance, suggesting that subspecies explain 8% of the total variance in microsatellite frequencies. However, this useful heuristic technique only evaluates a priori groupings, and the objective of the study ought to be to discover such groupings, not assume them. In fact, other arbitrary groupings of samples explained the same or greater amounts of variance, and I suggest that for testing subspecies limits, a gene tree is preferable. Grouping population samples by subspecies in the San Francisco Bay area accounts for 1.2% of the microsatellite variation, and despite claims that this informs conservation planning, the data do not support any particular population or subspecies as being genetically or evolutionarily significant. A distance phenogram was used to infer a sequential colonization of the Aleutian Islands, but because individuals were pooled into a priori groups and the phenogram was arbitrarily rooted, this conclusion is tenuous. A plot of heterozygosity vs number of alleles per sample showed that an equally parsimonious interpretation is that current genetic diversity tracks effective population size. Microsatellites should be replaced in nuclear-gene phylogeography by analyses of sequences, which will benefit the study of phylogeography, comparison of nuclear and mtDNA results, and aid in interpreting the results in a conservation context.

Phylogeography was originally introduced as a set of methods to discover recently isolated groups of individuals or populations (taxa) by superimposing a mitochondrial DNA (mtDNA) gene tree over geography (Avise et al. 1987). Phylogeography has subsequently expanded to include estimates of levels and patterns of gene flow, population growth, strength of selection, and times of divergence. From its beginnings, it has been acknowledged that sole reliance on mtDNA could lead to biases associated with the reliance on any single gene tree. Having abandoned allozymes a decade earlier, a search ensued for a set of variable nuclear loci that could complement, or perhaps replace, mtDNA. From several candidate techniques, analyses of microsatellite loci have now become common in studies of phylogeography. This class of molecular markers has several putative advantages, the most obvious of which is that many independent and highly variable loci can be analyzed. This could be an improvement over mitochondrial DNA (mtDNA) surveys, which yield a single, maternally inherited gene tree that is taken to represent the organismal or lineage history. However, there are drawbacks with the use of microsatellite data in phylogeography (Brito and Edwards 2008), having to do with sampling, homoplasy, relatively long coalescence times, and importantly, the analysis of data. Many of these issues are the same ones that resulted in the abandonment of allozymes in favor of mtDNA sequence data for assessing population history. Although the field of nuclear gene phylogeography is moving towards using sequences (Lee and Edwards 2008), many studies of geographic variation using microsatellites have been published, and their results merit examination.

The song sparrow Melospiza melodia is a widely distributed breeding species across North America including parts of Mexico (Arcese et al. 2002). Body size and plumage coloration vary greatly, especially among populations distributed along the Pacific coast and through the Aleutian Islands. The magnitude of phenotypic differences makes understandable the attention paid to this species by subspecies taxonomists and those interested in phenotypic variation (Aldrich 1984). The 30 or so described subspecies make it one of the most polytypic species in North America. The extensive geographic variation in plumage and body size has also attracted the attention of researchers interested in molecular phylogeography. Both mtDNA (Zink and Dittmann 1993, Fry and Zink 1998) and microsatellite studies (Chan and Arcese 2002, Pruett and Winker 2005, Pruett et al. 2008a,b) have been published.

In this paper, I reanalyze microsatellite data that were used to suggest that western North American song sparrow subspecies were at least in part genetically distinct (Pruett et al. 2008a, Chan and Arcese 2002). Secondly, I re-examine a conclusion reached by Pruett and Winker (2005) that microsatellite loci support a hypothesis of a linear, westward colonization of the Aleutian Islands. I comment on particular issues that I believe compromise the use of microsatellites in phylogeographic analysis.

Subspecies, detecting geographic differentiation, and molecular markers

Ornithologists have described a relatively large number of subspecies. However, it is generally agreed (Rising 2001, Remsen 2005) that many, if not the majority of named subspecies represent arbitrary geographic divisions of single character clines. Although these character clines might well reflect local adaptation, it is unlikely that all characters respond to geographically concordant selection gradients. If subspecies are based on one or a few traits that track geographically inconsistent selective gradients then subspecies will not predict overall patterns of character variation. Often when multiple phenotypic characters are analyzed in concert, subspecies boundaries are not apparent (Rising 2001). Furthermore, if the morphological traits that define subspecies are not hierarchically structured owing to not sharing a common evolutionary history, then subspecies are not coherent independent historical entities. These observations would explain why molecular markers often do not reflect subspecies limits (Zink 2004), as they track historical population subdivisions and not idiosyncratic selective gradients. However, many mtDNA studies detect major geographic subdivisions not predicted by subspecies limits, although they often correspond to groups of subspecies that have no formal taxonomic names or status (Zink 2004). Thus subspecies are the only taxonomic rank that is not consistent with evolutionary history. This is unfortunate because some governmental agencies must consider named subspecies as valid, evolutionarily distinct taxonomic units simply because they exist in official checklists, thereby misinforming conservation efforts.

Many authors, however, believe that microsatellites will corroborate subspecies limits where mtDNA has not, because the former set of markers has a high mutation rate and multiple loci can be analyzed. Unfortunately, basic coalescence theory predicts that if a population or group of populations are not reciprocally monophyletic on an mtDNA gene tree, neither will they be on a nuclear gene tree; this is a common empirical result (Zink and Barrowclough 2008). This result obtains owing to the larger effective population size of nuclear microsatellite loci and their concomitant longer coalescence times. If the amount of time since isolation of populations is less than Ne (2Nef for effective size of the female populations), then a nuclear locus, such as a microsatellite locus, with a coalescence time of 4Ne generations will have an exceedingly low probability of being reciprocally monophyletic (Hudson and Coyne 2002).

Microsatellite loci are initially screened and then the most variable ones are chosen for subsequent analysis (unlike allozyme analyses where monomorphic loci were reported). On average, microsatellite loci are far less variable than those chosen for inclusion in published studies (and if one sequenced the entire mitochondrial genome, it is likely that each individual would have a unique mtDNA haplotype). These chosen loci have high mutation rates, which some have incorrectly equated with rapid evolution. However, coalescence time is independent of mutation rate. Having more alleles means there are more twigs at the end of the branches (i.e., more resolution), not that these loci have an increased probability of detecting divergence. Zink and Barrowclough (2008) cited many studies that reported a shallow (and hence recent), but geographically structured mtDNA tree and a corresponding absence of structure in microsatellite data - exactly as predicted by basic coalescence theory. However, many of these studies attributed the lack of microsatellite differentiation to greater dispersal distances in males, whereas insufficient time since isolation is the most parsimonious interpretation.

Further compromising the use of microsatellites in phylogeography are the methods of analysis designed to detect geographic structure, which I argue is the first goal of phylogeography. Although the popular analysis of molecular variance (AMOVA) reveals the extent of population subdivision and is a useful heuristic measure, it does not reveal the geographic limits of historically independent groups nor their phylogenetic relationships. An ordination technique such as principal components analysis (PCA) can show the broad patterns of population structure in allele frequency space, but only in a general non-phylogenetic sense. The topology of a gene tree is better suited for this purpose. One often sees microsatellite studies in which individuals are grouped into samples based on geographic proximity, allele frequencies calculated for groups, and a phenogram derived from partitioning the matrix of pairwise genetic distances. This procedure (as well as AMOVA and PCA) is flawed owing to the assumption that individuals in a group are genealogically closer to each other than to those in other groups. That is, the goal of the study should be to discover genetic groupings, not assume them a priori. Some analyses such as STRUCTURE (Pritchard et al. 2000) use the individual as the unit of analysis, but the results are not easily interpretable in a phylogenetic sense, as they contain no information about hierarchal patterns (see Tishkoff et al. [2009] for an attempt to deduce phylogenetic information from a STRUCTURE analysis), and there is no root (see below). Additionally, it is likely that allele phenotypes with the same repeat number that occur in geographically disjunct localities are convergent and not homologous (Zink 2008). Thus the inability to produce a valid phylogenetic analysis reveals why microsatellites are poorly suited to achieve the goals of phylogeography, and are analogous to the reasons why gene tree approaches based on sequence data replaced allozymes, and will in my opinion shortly replace microsatellites.

For the initial process of discovery of phylogeographic pattern, mtDNA will be a primary tool. Nuclear loci, whether microsatellites or single gene sequences, and as noted above will be less apt to find geographic structure if the time since isolation is relatively short. However, to estimate confidence limits on parameters such as times of divergence, patterns of population expansion, effects of natural selection, or levels of gene flow, nuclear loci are needed, and sequences will be better suited than microsatellites (Brito and Edwards 2008). Microsatellites, however, will continue to be useful for the assessment of parentage, patterns of heterozygosity, and perhaps very local patterns of gene flow (based on the distribution of rare alleles).

Microsatellites and song sparrows

Pruett et al. (2008a) examined the relationship between microsatellite allele frequencies and subspecies boundaries in song sparrows distributed coastally from southern California to the Aleutian Islands. These authors (pp. 360) defined subspecies as "a collection of populations in a given geographic range that differ in some fixed way (almost always phenotypically) from other populations but that are not reproductively isolated from one another." If this were true, they would have been unable to test subspecies limits in song sparrows because there is no published documentation that any song sparrow subspecies fits this definition (Patten and Pruett 2009). Nonetheless, Pruett et al. (2008a) obtained data on 576 individuals from 23 localities (Fig. 1) representing 13 named subspecies (for other uses of the same data see Pruett et al. 2008b). The seven microsatellite loci analyzed revealed considerable variation (as expected). To examine the geographic structure of this variation, three hierarchical analyses of molecular variance (AMOVA) were carried out. One model ("ALL") grouped population samples into subspecies, one evaluated only samples and subspecies from Alaska, and the third considered only populations occurring outside of Alaska. The total percentage of allelic variation distributed among samples in the ALL model was 11%, with 89% of the variation occurring within samples. The question of interest is how much of the 11% is explained by grouping samples into subspecies, which as Pruett et al. (2008a) showed is 8% (with 3% explained by variation among samples within subspecies). Apparently Pruett et al. (2008a) concluded that their data did not fit a step-wise mutation model, and hence did not compute RST (Slatkin 1995), which potentially accounts for the genetic distances among alleles. Using the same data as Pruett et al. (2008a), kindly provided by C. Pruett, I performed an AMOVA using Arlequin (Excoffier et al. 2005), finding that the RST -values in the ALL model are 4.0% among subspecies and 3.4% among populations within subspecies. Thus over 90% of the variance in microsatellite allele frequencies is unexplained by subspecies limits.

It is not obvious how much variation should be explained by an AMOVA before considering subspecies limits to be biologically significant (Hedrick 1999). FST ranges from zero to one (in theory), meaning that there is not an unambiguous cutoff point. One could rely simply on statistical significance, but with sufficiently large samples, FST values of 1% can be statistically significant, whereas their biological significance is dubious (Björklund and Berget 2009). Another concern is that an AMOVA is not designed to discover groupings that maximize the among-group variance. Also one cannot tell from an AMOVA whether differentiation is apportioned evenly among subspecies or whether one particularly divergent sample causes the significant FST value. This is where sequence data are better suited for assessing taxonomic limits because the topology of a gene tree(s) can be overlain on geography to determine which populations are historically distinct and to discover their hierarchical relationships - the essence of phylogeography.

To explore potential inferences from AMOVA, I performed an AMOVA on the Pruett et al. (2008a) data in which the samples from Attu, Adak, Alaska Peninsula and Kodiak Island, the outlying samples in their principal coordinates plot, were omitted; owing to missing data, two loci were excluded. The amount of variance among samples was 4.5%, of which 3.2% is explained by subspecies. Thus, a relatively large contribution to the total allelic differentiation is accounted for by the four excluded samples from Alaska and the Aleutian Islands. Comparing these four Alaskan samples versus all others (i.e., two groups) returned a value of 12.8% among groups and 6.0% among samples within groups. Therefore, the way in which samples are grouped influences the amount of variance explained. In addition, the grouping of samples by subspecies does not consider potential isolation-by-distance effects (see below).

Pruett et al. (2008a: p. 359) did not claim that their analysis supported all subspecies, only "that in some, but not all, instances neutral genetic structure corresponded to recognized phenotypic structure." Assuming that "recognized phenotypic structure" corresponds to subspecies limits, all that this means is that current subspecies explain a statistically significant portion of variance in allelic frequencies, not that the groupings have evolutionary or taxonomic significance, or more importantly, that they comprise in any way an optimal partitioning of genetic variance. In fact, based on the two-group AMOVA presented above, one could just as easily conclude that the microsatellite data supported two subspecies, not 13 (their STRUCTURE analysis suggested that 12 was the appropriate number, although whether this is statistically different from another number is unknown). Pruett et al. (2008a: p. 363) concluded that "Ten of the 13 subspecies that we examined showed some level of concordance between genotype and phenotype." This is misleading given that "some level of concordance" is undefined, and the statement would be true no matter which of a large number of possible groupings of samples were used. This misleading interpretation could be averted with a rooted gene tree.

Subspecies, microsatellites and conservation

Subspecies are often used in conservation planning. Several subspecies of song sparrow in central California are listed as subspecies of special concern by the state of California (http://www.dfg.ca.gov/biogeodata/cnddb/pdfs/SPAnimals.pdf ). Chan and Arcese (2002) surveyed nine microsatellite loci in the same specimens from the San Francisco Bay area used by Pruett et al. (2008a). Chan and Arcese (2002) determined that the subspecies were not distinct, but concluded that the populations could qualify as management units (sensu Moritz 1994) based on a significant FST value. I performed an AMOVA on the nine samples from this region (Fig. 1) using the data of Pruett et al. (2008a), which revealed an overall FST of 2.1%, of which 1.2% was among subspecies (Chan and Arcese [2002] reported values of 2.56% and 1.38%, respectively). Hence, there is no support for subspecies in this localized region, as Chan and Arcese (2002) noted. However, it is unclear how these data could be interpreted to mean that the local populations qualified as management units, given that it is unknown how the 1.2% is apportioned geographically among the nine samples (and there is no gene tree). In effect, Chan and Arcese (2002) suggested a conservation plan designed to protect 1.2 - 1.38% of the variance at neutral loci. I believe that this is an untenable position, especially given U.S. Congressional direction that Distinct Population Segments (analogous to management units) were to be listed "sparingly" under the Endangered Species Act. If we consider an FST-value of 1.2% to mean that the population groupings merit preservation, thousands of arbitrarily defined (and mutually overlapping) units would qualify for protection (see Hedrick 1999, Bjorklund and Bergek 2009). This would be a difficult position to defend, given limited conservation resources. The pattern of reciprocal monophyly on a DNA sequence-based gene tree provides a more objective way to identify significant historical units for conservation purposes (Zink 2004), and would not likely reveal any for the song sparrows of this local region.

Microsatellites and MtDNA

Pruett et al. (2008a:363) commented that "We found substantially more variance among subspecies, even within the non-Alaska populations, than has been reported using mtDNA sequence data alone (Fry and Zink 1998)." It is true that the amount of variance in mtDNA sequences distributed among subspecies was effectively zero, compared to the 8% suggested by the AMOVA reported by Pruett et al. (2008a). However, 27% of the variance in the mtDNA sequences is distributed among populations within subspecies (Fry and Zink 1998). Thus, the mtDNA data actually revealed more structure than the microsatellites, it was just not aligned with existing subspecies limits. Also, Fry and Zink's sampling (1998) included individuals from North Carolina, Newfoundland, southern California and Alaska, a much wider range than that studied by Pruett et al. (2008a), and one might expect greater differentiation because of the larger area surveyed.

Because of differences in effective population size between the two markers, one cannot directly compare FST-values for mtDNA and microsatellites. However, Brito (2007) noted that we can predict the nuclear value as: Fst-nuc = Fst-mt/(4 - 3Fst-mt). Using the mtDNA value of 0.27 yields a predicted FST for nuclear loci of 0.08. Although the two studies covered vastly different amounts of the range, the predicted value is close to that obtained from microsatellites (0.11). Hence, it is not obvious that microsatellites revealed more differentiation.

Evolutionary history of song sparrows in the Aleutian Islands

Song sparrow size and plumage sootiness reach their maximum extent in the two terminal Aleutian Islands, Attu and Adak. Pruett and Winker (2005:1428) stated "The population genetics of northwestern song sparrows appear to fit a linear stepping-stone colonization model (Le Corre and Kremer 1998) from southeast to northwest." This conclusion was based on a genetic distance phenogram which showed (assumed) populations from Adak and Attu islands as sisters, with progressively chained population samples from Alaska Peninsula, Kodiak Island, Copper River delta, and Hyder (see Fig. 2). Two other samples (Queen Charlotte Islands and Alexander Archipelago) were grouped together, and were sister to these aforementioned samples. Pruett and Winker (2005) interpreted the sequence of samples in the distance phenogram as indicating a sequential northwestward colonization, terminating in Attu and Adak. Given that this is the extreme of the species' linear distribution in the Aleutians, this colonization scenario makes sense.

There is, however, a problem with the interpretation of the phenogram. The phenogram was midpoint rooted (or rooted at the first input taxon), hence lacking evolutionary direction. Thus, one could just as easily infer that the Queen Charlotte Islands and Alexander Archipelago were colonized from the Aleutian Islands and southeastern Alaska. That is the Achilles Heel of unrooted distance trees. Pruett et al. (2008a) do not specifically comment on the possible routes of colonization of song sparrows in the Aleutian Islands, but they had the data to do so. I converted their microsatellite allele frequencies into various distance measures in ARLEQUIN (FST, Slatkin's (1995) distance, Reynolds et al.'s (1983) distance) and subjected them to various clustering algorithms (Neighbor joining, UPGMA) in MEGA4 (Tamura et al. 2007). Pruett and Winker (2005) stated that their results were robust to differing distance measures. The midpoint rooted tree (Fig. 2) based on Slatkin's distance derived from the samples in common between Pruett et al. (2008a) and Pruett and Winker (2005) has the same topology as that shown in Pruett and Winker (2005) and only differs in how the tree is arbitrarily rooted by the analysis. From Figure 2, one would not conclude that the Aleutians were colonized from coastal continental samples. Instead, one could conclude that song sparrows expanded their range south along the coast from Alaska and the Aleutian Islands. This might be the case if the Aleutians were in fact a refuge, as suggested for other birds (Pruett and Winker 2008).

If the pattern of northwestward colonization hypothesized by Pruett and Winker (2005) is correct, one ought to see its signal in the geographically expanded data set of Pruett et al. (2008a), as their samples include California, British Columbia and Alaska. The neighbor-joining tree of all 23 samples (not shown) and its bootstrapped consensus (Fig. 3) is consistent with the interpretation of Pruett and Winker (2005). However, the tree is also consistent with an alternative conclusion - southward colonization, as noted above. The main message is that there are as many interpretations as there are potential roots to a distance tree. One cannot interpret an arbitrarily rooted tree in an evolutionary context with a proper root, usually determined by referencing an outgroup. The sister group of the song sparrow consists of Lincoln's sparrow M. lincolnii and swamp sparrow M. georgiana. Although these species were not studied, it is likely that they are too distant from the song sparrow to share alleles that could aid in rooting. A sequence based tree can in theory and usually in practice be rooted with an outgroup sequence. Such a phylogenetic analysis of sequences would be required to verify the conclusions of Pruett and Winker (2005).

Pruett and Winker (2005) also used the pattern of genetic variability to infer the pattern of colonization. In their data set, they observed a westward decrease in heterozygosity, suggesting a history of leading edge expansion (Hewitt 2004). I computed observed heterozygosity for each sample, also finding a pattern of decreasing variation in the western Aleutians (Fig. 4). However, an alternative interpretation exists, namely that the westward decrease in heterozygosity simply reflects a pattern of decreasing effective population size. If populations are smaller in the Aleutians than they are in continental regions, which seems likely, this will be reflected in measures of genetic variability. A more specific prediction is possible. In the simple case of a bottleneck or leading edge expansion, one expects allelic diversity to decay more rapidly than heterozygosity, because one loses rare alleles disproportionately, but they have relatively little effect on heterozygosity. I plotted the geographic trends in both observed heterozygosity and number of alleles per sample (standardized to a value of 1), predicting that if Pruett and Winker (2005) were correct about the direction of colonization, the two trends would be different. Specifically, allelic diversity should decay more steeply than heterozygosity as one proceeded northwestward in the Aleutian chain. The results (Fig. 4) show that the two values covary remarkably consistently. Therefore an equally parsimonious interpretation is that populations on Adak and Attu are relatively small owing to island size and the harsh environmental conditions, resulting in lower genetic variation.

Another aspect of the distance phenogram (Fig. 3) merits attention. The topology suggests the existence of three groups of samples corresponding to California, British Columbia, and Alaska. It is clear from the geographic distribution of samples (Fig. 1) that the apparent genetic gaps simply reflect sampling gaps. If samples from intermediate areas were obtained, it is highly likely that the "gaps" would fill in, and the three groups would be an illusion. This is a serious concern for many phylogeographic studies irrespective of molecular marker.


In the song sparrow, as in other species (Zink and Barrowclough 2008, Barrowclough and Zink 2009), the use of microsatellites did not reveal a different view of population history over that inferred from mtDNA data. It is important to corroborate conclusions drawn from any set of molecular markers, but the point is that the mtDNA was not misleading or erroneous, as has been suggested (e.g., Edwards et al. 2007, Edwards and Bensch 2009). Subspecies of song sparrows are not supported by either class of molecular marker, either owing to their recent origin, or the fact that they are based on morphological characters responding to idiosyncratic selection gradients. The lack of rooting compromised many of the uses for which the microsatellite phenograms were used. It is my opinion that for these and other reasons (e.g. difficulty in dating divergence events), if mtDNA phylogeographic results are to be tested, it ought to be with sequences from nuclear loci (Lee and Edwards 2008) analyzed with robust coalescence methods. Researchers should recognize that for recently isolated groups, mtDNA will be the main way to recognize historical groupings of populations, and one should not expect nuclear loci to detect recent differentiation owing to their longer coalescence times. Multiple nuclear loci will be important for estimating parameters such as levels of gene flow and time since divergence (and their confidence intervals). Lastly, there is a striking historical parallel between allozymes and microsatellites. Studies of mtDNA sequences supplanted allozyme methodology because of the many drawbacks of allele-frequency based analyses. I predict that these same drawbacks will result in analyses of nuclear sequences replacing microsatellites in phylogeography.