Conservation of human microsatellites across 450 million years of evolution.

The sequencing and comparison of vertebrate genomes have enabled the identification of widely conserved genomic elements. Chief among these are genes and cis-regulatory regions, which are often under selective constraints that promote their retention in related organisms. The conservation of elements that either lack function or whose functions are yet to be ascribed has been relatively little investigated. In particular, microsatellites, a class of highly polymorphic repetitive sequences considered by most to be neutrally evolving junk DNA that is too labile to be maintained in distant species, have not been comprehensively studied in a comparative genomic framework. Here, we used the UCSC alignment of the human genome against those of 11 mammalian and five nonmammalian vertebrates to identify and examine the extent of conservation of human microsatellites in vertebrate genomes. Out of 696,016 microsatellites found in human sequences, 85.39% were conserved in at least one other species, whereas 28.65% and 5.98% were found in at least one and three nonprimate species, respectively. An exponential decline of microsatellite conservation with increasing evolutionary time, a comparable distribution of conserved versus nonconserved microsatellites in the human genome, and a positive correlation between microsatellite conservation and overall sequence conservation, all suggest that most microsatellites are only maintained in genomes by chance, although exceptionally conserved human microsatellites were also found in distant mammals and other vertebrates. Our findings provide the first comprehensive survey of microsatellite conservation across deep evolutionary timescales, in this case 450 Myr of vertebrate evolution, and provide new tools for the identification of functional conserved microsatellites, the development of cross-species microsatellite markers and the study of microsatellite evolution above the species level.

Introduction

Microsatellites are arrays of short, tandemly repeated, DNA motifs (1–6 bp) found throughout the genomes of both prokaryotes and eukaryotes (Buschiazzo and Gemmell 2006). Their distribution and density in genomes appear to be nonrandom but can vary greatly even between closely related species (Tóth et al. 2000; Warren et al. 2008). Polymorphism at microsatellite loci occurs through additions and deletions of motifs in the repeat array, but the detailed dynamics of microsatellite mutation are still not fully understood (Buschiazzo and Gemmell 2006). Microsatellites have gained notoriety in medical genetics with evidence of association with colorectal, endometrial, and various other cancers (Woerner et al. 2006), and the implication of unstable repeats in ∼30 human hereditary disorders (Mirkin 2007). Other microsatellites, in contrast, are thought to play an advantageous role in evolution (Kashi and King 2006; Vinces et al. 2009). However, microsatellites have attracted the widest interest as polymorphic, neutral genetic markers for population genetics, gene mapping, forensics, or paternal investigation (Schlötterer 2004).

Being traditionally regarded as neutrally evolving, nonfunctional and highly polymorphic sequences, microsatellites are not expected to be retained in different species, particularly when evolutionary distance increases. This view is supported by the relative difficulty to transfer microsatellite markers between distant species, which severely limits cross-species applications in molecular ecology (Barbara et al. 2007). However, not only have these assumptions never been thoroughly tested, they run counter to theoretical expectations (Tachida and Iizuka 1992; Stephan and Kim 1998) and direct observation of microsatellite conservation in closely related species (Schlötterer et al. 1991; Blanquer-Maumont and Crouauroy 1995; Primmer et al. 1996; Gemmell et al. 1997; Crawford et al. 1998; Slate et al. 1998; Guillemaud et al. 2000; González-Martínez et al. 2004), as well as species that diverged 100+ MYA (FitzSimmons et al. 1995; Rico et al. 1996; Ezenwa et al. 1998; Moore et al. 1998). Unfortunately, these reports have thus far been limited to one or few loci, whereas genomewide searches for homologous human microsatellites have to date been limited to comparisons with chimpanzee (Kayser et al. 2006; Vowles and Amos 2006; Kelkar et al. 2008) or other close primate relatives (Raveendran et al. 2006). In response to the lack of both comprehensiveness and evolutionary scope in prior analyses of microsatellite conservation, it is timely to develop a reliable method to identify, at the genome scale, human microsatellites conserved in mammals and beyond.

Recently, genome sequence projects have dramatically increased in number and evolutionary breadth, providing the opportunity to advance our understanding of the organization, evolution, and functional landscape of eukaryotic genomes, as was emphasized by the recent findings of the ENCODE Project Consortium (Gerstein et al. 2007; King et al. 2007; The ENCODE Project Consortium 2007; Thurman et al. 2007). In particular, comparative methods have been developed to predict evolutionarily conserved and/or functional sequences (Margulies and Birney 2008) not only among mammalian genomes (Waterston et al. 2002; Cooper et al. 2005; Lindblad-Toh et al. 2005; Mikkelsen et al. 2007) but also among more distant vertebrates, including avian, amphibian, and fish species (Siepel et al. 2005; Venkatesh et al. 2006; Loots and Ovcharenko 2007). The comparative method of choice to identify human conserved elements relies on the statistical prediction of constrained segments in pairwise and multiple sequence alignments (Cooper et al. 2005; Siepel et al. 2005; Prabhakar et al. 2006); but is this method applicable to any type of DNA sequence? Unfortunately, using current algorithms, microsatellite sequences do not align well and, at first sight, might resemble sequences with no common ancestry; the above statistical approach, which assumes a “perfect” alignment (Margulies and Birney 2008), is thus inappropriate for microsatellites. An alternative approach, applied in recent studies of human–chimpanzee comparisons (Kayser et al. 2006; Vowles and Amos 2006; Kelkar et al. 2008), is to identify all microsatellites in each genome and find homologies by comparing positions in a pairwise whole-genome alignment. Although efficient, this task may become impractical when many genomes are compared and probably disproportionate when dealing with highly divergent species that are not expected to share many microsatellite sequences (e.g., human and chicken). We instead sought to narrow our investigation down to a subset of genomic sequences already aligned to each other and thus likely to contain the subset of conserved microsatellites. The publicly available alignment of the human genome against 16 vertebrate genomes, namely, 17-way alignment (17-WA), provided a timely framework to investigate the extent and patterns of conservation of human microsatellites.

Our proposed approach has already been briefly introduced in the recent analysis of the platypus genome using an alignment of six vertebrate genomes (Warren et al. 2008), but here we present the detailed methodology and analysis of human microsatellite conservation across vertebrate genomes representing 450 Myr of evolution. This work offers clues to explain the wide conservation of microsatellites, assesses the reliability of our findings, and provides the first opportunity to discuss the implications and applications of conserved microsatellites in evolutionary genomics and genetics. Indeed, in addition to providing new lines to explain genome organization and evolution and the function of microsatellites within genomes, our efforts will improve the prospects of transferring microsatellite markers between related species to promote comparative gene mapping (Sun and Kirkpatrick 1996), cutting the development costs of de novo microsatellites (Barbara et al. 2007), and the opportunity to study microsatellite evolution above the species level (Zhu et al. 2000; Kelkar et al. 2008).

Materials and Methods

Vertebrate Sequences

The 17-WA available on the University of California in Santa Cruz (UCSC) Genome Browser for each human chromosome was downloaded by anonymous FTP from ftp://hgdownload.cse.ucsc.edu/goldenPath/hg18/multiz17way/. Multiple alignment format blocks were extracted and converted to FASTA format using a stand-alone version of Galaxy (Giardine et al. 2005). Due to the large size of the alignments for chromosomes 1–4, the files were split in half; this had no consequence except that an additional step to merge results for each respective chromosome was required. Sequence gaps were removed using the degapseq module from the EMBOSS 5.0 package (Rice et al. 2000).

Microsatellite Search and Classification

Our approach aimed at 1) using a fast, flexible, reproducible, and user-friendly program and 2) finding perfect and imperfect microsatellites, with a repeating motif of size 1–6 bp and no shorter than 12 bp for mono-, di-, tri- and tetranucleotide repeats and three perfect repeats for penta- and hexanucleotide repeats. Perfect and imperfect microsatellites (motif length: 1–6 bp) were searched in ungapped sequences using SciRoKo 3.1 (Kofler et al. 2007) with fixed penalty parameters (score: 12, mismatch penalty: 4, simple sequence repeat [SSR] seed minimum length: 3, SSR seed minimum repeats: 3, maximum mismatches at once: 3).

Genomic intervals of microsatellites in each vertebrate genome were recorded with block number, standardized repeat motif (Kofler et al. 2007), array length, and number of imperfections. Microsatellites in the alignment of the Y chromosome were not included in analyses because only human, chimp, and mouse Y chromosomes are included in the 17-WA (but see supplementary table S3, Supplementary Material online). Human microsatellites lying in segmental duplications >1 kb and >90% identity (Bailey et al. 2001) and nonhuman overlapping intervals (5 bp minimum cutoff), indicating those sites that aligned to human duplicated segments, were removed. Indeed, every alignment block in the 17-WA represents one, and only one, human interval, but the same nonhuman interval can be assigned to one or more human intervals. Intervals overlapping with repeats other than simple repeats or low-complexity sequence (Smit et al. unpublished data) were also discarded. Segmental duplication and repeat data were retrieved from the UCSC Table Browser (Karolchik et al. 2003).

We classified microsatellites as simple, compound, linked, and mixed loci. If the sequence 25 bp upstream and downstream of a microsatellite interval did not contain another microsatellite, the microsatellite was classified as “simple.” Microsatellite segments were merged and classified as “compound” if they were 5 bp or less apart from each other or were overlapping by 5 bp or less, “linked” if they were separated by 5–25 bp, or “mixed” if they contained both linked and compound portions. This classification is necessary because complex structures ought to be considered as individual microsatellites rather than several independent loci: 1) complex microsatellites tend to evolve differently (Buschiazzo and Gemmell 2006; Kofler et al. 2008), 2) a considerable fraction of microsatellites (3–25%) are part of a compound structure in vertebrate genomes (Kofler et al. 2008), and 3) to ensure that two neighboring microsatellites were separated by at least 25 bp of “unique” sequence, a sufficient length to design a potential primer for future comparative polymerase chain reaction (PCR)-based analysis. This series of operations produced, in human, a data set of 696,016 microsatellites covering 19.5 Mb of the human genome (0.70% of human sequences in the 17-WA).

Microsatellite Conservation

Positions of nonhuman microsatellites were converted to the hg18 human assembly using the liftOver utility and chain files (Kent et al. 2003) available at the UCSC Genome Browser (Karolchik et al. 2003). Converted intervals overlapping with human repeats other than simple or low-complexity repeats were discarded. The fraction of human microsatellites overlapping with any of the converted microsatellite positions indicated conserved sites. We found 594,340 human microsatellites conserved in at least one species, that is, 85.0% of the initial data set.

G + C composition

We classified microsatellites conserved in at least one nonprimate species according to the G + C composition of their standardized motif as given in SciRoKo’s output (Kofler et al. 2007). G + C-rich motifs were characterized by a G + C content >50%, whereas A + T-rich motifs and AT = GC motifs were characterized by a G + C content <50% and equal to 50%, respectively. For practicality, repeat segments forming compound, linked, and mixed microsatellites were treated as individual microsatellites for this analysis.

Genomic Location

A tentative canonical list of 17,260 nonoverlapping human nuclear genes was produced from the UCSC Genome Browser and used to locate human microsatellites conserved in coding exons, 3′-untranslated regions (UTRs), 5′-UTRs, introns, or intergenic regions (IGRs). Conserved microsatellites spanning more than one element were positioned in the element with the longest overlap. When an equal overlap existed, we positioned the microsatellite following the preferential order given above.

Statistical Analyses

Genomic features were based on annotations of human autosomes obtained from the UCSC Genome Browser and were calculated in 1 Mb windows using Galaxy. Densities of microsatellites were based on sequence length excluding segmental duplications and repeats, unless stated otherwise. Windows with low-sequence coverage and high content of repeats and segmental duplications were excluded (i.e., windows with >70% of their length annotated as gaps and segmental duplications and windows with >90% of their length annotated as gaps, segmental duplications, and repeats). Again, these repeats do not include low complexity or simple repeats. This treatment excluded 233 windows out of 2,857. We considered smaller window sizes (500 kb and 250 kb) but selected 1 Mb windows as only a negligible number of these contained no microsatellite conserved in at least three nonprimate species (23 out of 2,624 windows). Spearman’s rank-order correlation tests were performed using the R package (www.r-project.org).

Method Assessment

To assess the validity of the identified conserved microsatellites, we compared their positions with regions previously found to be suspiciously aligned in the 17-WA of human chromosome 1 using a statistical assessment (Prakash and Tompa 2007). Any conserved microsatellite overlapping with regions identified as suspiciously aligned may be considered suspicious too.

Results

Alignability to the Human Genome

The UCSC team produced the 17-WA blocks using the human genome as reference (supplementary table S1, Supplementary Material online); the end result is therefore not an all-against-all genome alignment. For this reason, our results ought to be presented from the human perspective too. Besides, the organization of the human genome and its alignability to other genomes ought to be made explicit to fully comprehend our results.

The alignable and unique fraction of the human genome, that is, the fraction studied here, represented only ∼37% of its total length and was fairly heterogeneous between chromosomes (supplementary table S2, Supplementary Material online). Chromosomes 18 and 13 were highly represented in the 17-WA (42.26% and 41.72%, respectively), whereas X, 19, 22, and 16 had the lowest representations (28.42%, 30.14%, 33.43%, and 34.17%, respectively). The origin of this disparity is essentially interchromosomal differences in 1) the amount of gaps in the sequence, mostly a result of high heterochromatin content, 2) content of segmental duplications and repeats, 3) other genomic features that may affect microsatellite distribution, for example, gene density, and 4) sequence alignability with other genomes (supplementary fig. S1, Supplementary Material online).

As expected, there was a negative relationship between the size of sequence aligned to the human genome (alignability) and both the phylogenetic distance from human to each comparison species and times of divergence from the common ancestor (supplementary table S1, Supplementary Material online), and this relationship was found to be best explained by an exponential decline (R2 = 0.9581 and R2 = 0.9625, respectively). We nevertheless found large differences in alignability between species more closely related to one another than to human. than to human. For example, the relatively low amount of human sequence aligned to the mouse genome (∼37%) compared with the dog and cow genome (∼57% and 51%, respectively) reflects the well-known higher rate of sequence evolution and large deletions occurring in the rodent lineage, probably a reflection of short generation time (Waterston et al. 2002; Lindblad-Toh et al. 2005).

Microsatellite Survey

Mining microsatellites in genomic sequences is not a trivial task; the chosen approach and resulting data set largely depend on the underlying objectives of the research (Merkel and Gemmell 2008). Our methodology was developed to look for all orthologous microsatellites for which PCR primers could be potentially designed in supposedly unique genomic sequences for downstream cross-species applications, for example, comparative mapping, population genetics of nonmodel species, and study of interspecies microsatellite evolution.

Our initial set of human microsatellites (HMs) comprised a total of 696,016 microsatellites (supplementary tables S2 and S3, Supplementary Material online), including 11.35% with complex structures (compound, linked, or mixed microsatellites). Based on total ungapped length of chromosomes, HM density appeared particularly homogeneous among human autosomes (249.3 ± 10.6 HM/Mb) but the X chromosome exhibited a lower density (207.1 HM/Mb). Conversely, when densities were based on the length of ungapped sequences free of segmental duplications and repeats to account for the previously mentioned differential representation of each chromosome in the unique fraction of the alignment, the overall picture was comparably heterogeneous (495.7 ± 37.2 HM/Mb): chromosomes 19, 16, 20, X, and 22 showed a relative increase in HM density (639.18; 550.47; 512.55; 535.14; and 508.92 HM/Mb, respectively), whereas chromosomes 13 and 18 showed a slight decrease (470.54 and 474.41 HM/Mb, respectively). This second measure of density compared favorably with chromosomal differences found in a genomewide scan of human microsatellites (Subramanian et al. 2003) and confirmed that the HM data set, which we refer to as our background distribution, represented well the overall distribution of microsatellites in the human genome.

We compared microsatellite abundance in every genome relative to that of HMs (fig. 1 and supplementary table S2, Supplementary Material online) and found proportions ranging from 87.88% in chimpanzee to 15.52% in opossum for mammals and to 1.24% in fugu for vertebrates. These results are positively correlated with the amount of sequence aligned in each species (Spearman’s rank correlation, ρ = 0.89, P < 0.0001) and are thus also dependent on phylogenetic distance from human.

F

Species-specific microsatellite enrichment. (A) Alignability to the human genome and conservation of human microsatellites in vertebrate species. (B) Scatter plot showing the ratio (rp) of percentage of microsatellite conservation to percentage of alignment relative to human. Dotted lines represent a 5% significance threshold. Species are arranged from left to right by increasing distance (substitution rate) from human (Miller et al. 2007).

By measuring the ratio of the percentage of microsatellite abundance to the percentage of human sequence that aligns to each genome, we also obtained an indication of whether sequences from each species were enriched or impoverished for microsatellites compared with human sequences (fig. 1). Rather than following a phylogenetic trend, this ratio demonstrated species-specific enrichment. Microsatellites were especially enriched in mouse, elephant, and dog, whereas sequences from armadillo, frog, and fugu were particularly depleted in microsatellites in comparison with the human genome. These differences may be caused by species-specific microsatellite birth and death events (Buschiazzo and Gemmell 2006) and/or by the species-specific nature of alignable sequences. Our enrichment results are concordant with independent analyses of whole-genome microsatellite coverage in mouse, dog, opossum, and chicken (Waterston et al. 2002; Warren et al. 2008), thus we would favor the former hypothesis.

Phylogenetic Extent of Conserved Human Microsatellites in Vertebrate Genomes

We define our conserved microsatellites as single-copy, orthologous arrays of short tandem repeats, regardless of the specific nature of their primary sequence, that are found in genomic regions that similar to the human genome in one or several species that they could be aligned with the BlastZ/MULTIZ algorithms (Schwartz et al. 2003; Blanchette et al. 2004) used to construct the UCSC 17-WA (supplementary fig. S2, Supplementary Material online).

Of 696,016 microsatellites identified in human aligned sequences, 594,340 (85.39%) were found to be conserved in at least one comparison species, whereas 199,403 (28.65%) and 41,608 (5.98%) were conserved in at least one and three nonprimate species, respectively (supplementary table S3, Supplementary Material online). The fraction of conserved human microsatellites decreased from 87.74% (521,476) in chimpanzee to 1.71% (10,140) in opossum for mammals and to 0.16% (961) in fugu for vertebrates (supplementary table S2, Supplementary Material online). These results demonstrate a much higher extent of microsatellite conservation than previously found in mammals (Moore et al. 1998) but still illustrate an overall dramatic decline of human microsatellite conservation across vertebrates. In fact, we found that microsatellite conservation decayed exponentially with increasing phylogenetic distance from human (fig. 2). It can be argued that measuring the fraction of conserved human microsatellites using absolute numbers of microsatellites may be biased by the amount of aligned sequences, which were shown above to decline exponentially over time. Substituting microsatellite densities for raw numbers of microsatellites may well resolve this potential issue. However, if microsatellite conservation strictly follows alignability, and based on the analysis of the 28-WA (Miller et al. 2007), we would expect a much slower rate of exponential decay of microsatellite conservation in coding sequences than that pictured in figure 2. Therefore, we believe that our use of absolute numbers, while inducing a potential bias, likely has little impact on the interpretation of the data.

F

Phylogenetic extent of conservation of human microsatellites. (A) Decay of conservation in different genomic locations as a function of phylogenetic distance from human (Miller et al. 2007). Microsatellite conservation is measured as the fraction of human microsatellites identified in the aligned portion of the human genome that is found conserved in at least one other species. Scatter plots for total microsatellites and microsatellites in introns and intergenic regions (IGR) overlay. (B) Conservation profiles of human microsatellites in vertebrate genomes. Each profile is a proportional distribution of the range of conservation of human microsatellites conserved in at least each of the species, from exclusive (1 species, leftmost bar) to wide (12 species, rightmost bar). Each bar thus represents a percentage of human microsatellites that fall in each range category, for each species, and bars of identical range add up to 100%. No microsatellite was found in 13 species and only one in all 14 species. Species are arranged from left to right by increasing branch length from human (Miller et al. 2007). Primates were excluded to allow the observation of differences among species distantly related to human.

To explore patterns of microsatellite conservation further, we examined the proportion of human microsatellites conserved within species subsets (fig. 2). A profile skewed to the left indicates a species that shares microsatellites relatively exclusively with human, whereas a profile skewed to the right indicates a species that mostly shares microsatellites that are broadly conserved. Under a neutral model of evolution, these scenarios would be typical of species that are respectively closer (e.g., dog) and more distant (e.g., zebrafish) to human. Figure 2 shows that this expectation is in relatively good agreement with our observations, with intermediate stages between the two extremes. In fact, only species with a mere 2X coverage did not perfectly fit with this general pattern, for example, armadillo, the closest species to human if 4-fold degenerate site substitutions are used to measure phylogenetic distance, revealed a flat conservation profile instead of the expected skew to the left. We believe, however, that profiles from 2X covered genomes are not complete and that any premature interpretation should therefore be avoided.

Interchromosomal Distribution of Human Conserved Microsatellites

We sought to investigate whether there was any pattern in the distribution and extent of microsatellite conservation at the chromosome level by counting microsatellites conserved in increasing number of species. As expected, there was a rapid decline of conserved microsatellites with increasing species number, regardless of the chromosome examined (fig. 3).

F

Distribution of human microsatellites conserved in nonprimates species. The number of species is color coded as indicated in the legend.

We found it pertinent to compare HMs abundance with numbers of microsatellites conserved in 1) at least one of all 16 species (human conserved microsatellites [HCMs]), 2) at least one of the nonprimate species (NPMs), and 3) at least three of the nonprimate species (NP3Ms). At the genome scale, the three inclusive subsets represented 85.39%, 28.65%, and 5.98% of the initial data set, respectively (supplementary table S2 and supplementary fig. S3, Supplementary Material online).

At the chromosome level, proportions of HCMs compared with HMs were strikingly homogeneous (84.28–86.61%) with the exception of chromosomes 19, X, and 22 (77.10%, 78.86%, and 82.13%). Accordingly, the alignability of human chromosomes 19 and X was the lowest among eutherian genomes, especially primate genomes that contain most microsatellites in the HCM data set (66.45%) and thus influence greatly the overall HCM distribution (supplementary fig. S4, Supplementary Material online). When primate-specific microsatellites (PSMs) were excluded, proportions of NPMs were more heterogeneous among human chromosomes (25.44–30.68%), although chromosome 19 still showed a distinctively low proportion (22.07%). When NP3Ms only were considered, interchromosomal differences in the extent of human microsatellites were manifest (4.13–8.09%) and did not follow previous observations, for example, chromosome 19 had a comparatively average proportion of microsatellites conserved in at least three nonprimate species (5.71%). Yet again, these results might be caused by the uneven alignability of human chromosomes to other genomes: chromosome 19 was comparatively highly represented in species distant to human, that is, in opossum but especially in nonmammalian vertebrates (supplementary fig. S4, Supplementary Material online), a likely explanation for the relatively higher proportion of NP3M conservation. Chromosomes 1, 11, 15, 16, 17, and 22 also showed high representation in distant species and high proportion of NP3M, whereas chromosomes 4 and 13 showed the contrary dispositions.

Overall, our results of interchromosomal distribution of conserved microsatellites showed that the distribution of conserved microsatellites broadly corresponded to the overall distribution of aligned, hence conserved, genomic sequences and suggested that a finer scale analysis of microsatellite distribution in relation to other genomic elements, such as genes, would help understand why microsatellites in different chromosomes were differentially maintained in genomes. Indeed, it is striking that gene density is highest in highly aligned chromosomes, whereas chromosomes 4 and 13 share the lowest densities with chromosome 18. In addition, the former group contained proportionally more NPMs in exons than the latter group (supplementary fig. S5, Supplementary Material online).

Megabase Distribution of Human Microsatellite Conservation

We sought to inspect what could drive the distribution of conserved microsatellites at a finer scale than the chromosomes level, which may help understanding the causes of microsatellite conservation in genomes.

We first compared density of human microsatellites (HMs) in 1 Mb windows of autosomes with densities of human microsatellites conserved in at least one other species (HCMs), in primates only (PSMs), in nonprimate species (NPMs), and in at least three nonprimate species (NP3Ms). We found a general positive correlation between HM density and densities of all sets of conserved microsatellites (table 1 and supplementary fig. S6, Supplementary Material online), although the statistical significance was weaker for NPMs and especially NP3Ms, which suggests that a number of megabase segments contain a higher than usual proportion of widely conserved microsatellites.

Table 1

Covariation between Human Microsatellites and Other Genomic Features

	HM	G + C	Gene	SINE	LINE	LTR	Rrecomb	SNP	cIND	tfbs
HM	—	n.s.	−0.22***	0.11***	−0.37***	−0.26***	−0.33***	−0.05**	0.41***	0.40***
HCM	0.98***	−0.07***	−0.25***	0.14***	−0.32***	−0.24***	−0.31***	−0.09***	0.45***	0.42***
PSM	0.90***	−0.11***	−0.27**	0.17***	−0.28***	−0.12***	−0.26***	n.s.	0.16***	0.15***
NPM	0.84***	n.s.	−0.19***	0.08***	−0.27***	−0.29***	−0.28***	−0.16***	0.67***	0.61***
NP3M	0.63***	0.17***	0.04*	0.15***	−0.35***	−0.41***	0.25***	−0.23***	0.74***	0.75***
A + T-rich	—	−0.33***	−0.38***	0.24***	−0.04*	−0.14***	−0.13***	−0.21***	0.61***	0.48***
G + C-rich	—	0.65***	0.45***	0.51***	−0.62***	−0.54***	0.35***	−0.08***	0.41***	0.57***
AT = GC	—	n.s.	−0.24***	0.17***	−0.20***	−0.17***	−0.32***	−0.07***	0.54***	0.46***

NOTE. —Left to right: Density of microsatellites in aligned sequences (HM) and conserved in at least one species (HCM), primates only (PSM), and at least 1 (NPM) and 3 (NP3M) nonprimate species; NPMs are also differentiated as A + T-rich (motif G + C content <50%), G + C-rich (>50%), and AT = GC (=50%) (see Materials and Methods); G + C content; gene density; SINE, LINE and LTR coverage; average recombination rate; SNP density; indel-purified sequence coverage (cIND), and density of tfbsCons. Source: UCSC Genome Browser. Spearman’s rank correlation factor ρ, P value significance: 0 < *** < 0.001 < ** < 0.01 < * < 0.05 < not significant (n.s.).

We further carried out these comparisons relative to sequence composition (G + C content), genomic elements (gene density and repeat coverage), and four measures of evolutionary change; two derived from the human genome (recombination rate and single nucleotide polymorphism [SNP] density) and two derived from genomic comparisons (coverage in conserved, “indel-purified,” intervals, viz. cIND and density of conserved transcription factor binding sites, viz. tfbsCons). Preliminary correlation analyses between these factors confirmed results from previous analyses of the human genome (e.g., Fullerton et al. 2001; Lander et al. 2001): G + C content covaried positively with gene density, short interspersed repeat element (SINE) density, and recombination rate but was inversely correlated with long interspersed repeat element (LINE) and long terminal repeat (LTR) density (supplementary table S4, Supplementary Material online). Table 1 shows correlations between these genomic features and our microsatellite data sets. As a whole, microsatellite densities were negatively correlated with gene density, LINE, LTR, and recombination and weakly associated with SINE coverage. The strongest relationships (ρ > 0.40) appeared between microsatellites, including the background HM data set and measures of sequence conservation (cIND and tfbsCons, supplementary fig. S6, Supplementary Material online). A relatively strong relationship with the background distribution seems unexpected but can be explained by our biased alignment approach, that is, microsatellites were scanned in sequences known to possess some level of conservation; however, the much larger significance found with NPMs and NP3Ms (especially compared with PSMs) shows that widely conserved microsatellites were mostly found in the vicinity of other conserved sequences. This view is further supported by the negative relationship found between SNP density and both overall conservation and the most widely conserved set of microsatellites (NP3Ms).

To explore whether the extent of microsatellite conservation was affected by G + C composition, we grouped NPMs into G + C-rich, A + T-rich, and AT = GC microsatellites, depending on whether G + C content of the repeat motif was superior, inferior, or equal to 50%, respectively. G + C-rich NPMs were found to cluster in G + C-rich regions and were therefore typically, though weakly, associated with genes, SINEs, and high recombination rate and inversely correlated to LINE and LTR density (table 1). A + T-rich NPMs generally showed a contrary disposition and AT = GC NPMs generally showed an intermediate disposition, although they had a weaker association to SINEs and recombination rate than A + T-rich NPMs. Of the three data sets, A + T-rich NPMs had the strongest negative relationship with SNP density. At least for NPMs, it thus seemed that the G + C composition of conserved microsatellites correlated with that of the surrounding sequences. Due to low numbers and for statistical purposes, we did not partition NP3Ms relative to their G + C composition, but an overall small association with G + C-rich regions (table 1) and an analysis of microsatellite composition in the different data sets (supplementary fig. S7, Supplementary Material online) showed that A + T-rich microsatellites were depleted in NP3Ms compared with other data sets.

Genic Environment Influences Microsatellite Conservation

When we looked at the distribution of human microsatellites in coding exons, UTRs, introns, and IGRs that are conserved in each of the comparison species, the vast majority of conserved microsatellites lied in nonexonic regions (supplementary fig. S8, Supplementary Material online). The proportions of microsatellites found in each genomic region were fairly constant for microsatellites conserved in eutherians, with ∼55–60% lying in IGRs of the human genome, ∼35% in introns, and ∼5–10% in exons (UTRs and protein-coding sequence) but varied considerably for microsatellites conserved in more distant species. The decrease in conservation was slower in exonic than nonexonic regions when phylogenetic distance increased (fig. 2), a pattern similar to that of evolutionary conserved regions (ECRs, Loots and Ovcharenko 2007). However, whereas Loots and Ovcharenko (2007) observed that >75% of ECRs shared between human and nonmammalian vertebrates were in coding regions, we found that at most 35% of conserved microsatellites were in exonic regions (human–fugu comparison). Although this figure might be underestimated due to spurious alignments with distant vertebrates (see below), it was anticipated in light of a well-known distribution bias of microsatellites toward nonexonic regions of vertebrate genomes (Tóth et al. 2000).

Overall, though, conservation of microsatellites in coding exons declined more slowly than conservation of microsatellites in UTRs, which in turn declined more slowly than loci in introns and IGRs, as illustrated in figure 2.

The Reliability of Large-Scale Alignment and Microsatellite Data Mining

Our results are only as accurate and reliable as the sequence assemblies, the genomic alignments, and the microsatellite search algorithm.

First, concordant with our experience and preliminary tests (Merkel and Gemmell 2008), SciRoKo (Kofler et al. 2007) has recently been recognized as a highly performing tool to mine for perfect and imperfect microsatellites in genomic sequences (Sharma et al. 2007). Although tolerating the identification of rather short arrays, which could help document the concept of microsatellite life cycle (Buschiazzo and Gemmell 2006), our search parameters were purposely conservative regarding purity: imperfect microsatellites that maintained a clear repeat pattern were included, but low complexity DNA and overdegenerated repeat sequences were ignored with no need for additional filtering of spurious sequences.

Second, coverage and accuracy, that is, extent of sequence gaps and errors, of the genomic assemblies available at the time and used to produce the UCSC 17-WA are variable (supplementary table S1, Supplementary Material online). In particular, the alignment contains four mammalian genome assemblies with a 2X depth coverage, namely rabbit, armadillo, elephant, and tenrec, which may significantly increase the amount of false negatives in our results. According to the Lander and Waterman (1988) formula, a 2X assembly should include 87.5% of the bases in the genome and a 5X assembly 99.4% (Miller et al. 2007). Although high coverage of every genome would clearly be preferable, increasing the available branch length with low-coverage assemblies still considerably improves the accuracy of multiple genome alignments (Margulies et al. 2006; Wong et al. 2008) and of the identification of short conserved elements (Eddy 2005) and therefore improves our analysis.

The UCSC 17-way and chain alignments are the third and arguably the most critical (Wong et al. 2008) source of potential inaccuracies and missing data in our results. This is caused by 1) erroneous or missing genomic sequences (see above), 2) the methodological difficulties to produce a true alignment for sequences generated from highly diverged species (Kumar and Filipski 2007), and 3) the phylogenetic tree used to construct the 17-WA that differ slightly from the most recent understanding of evolutionary relationships between the compared species (Miller et al. 2007). Also, unlike the recently updated 28-way and 44-way alignments, the generation of the 17-WA did not include filtering of pairwise alignments based on synteny (for high-quality mammalian sequences) and reciprocal best alignments (for 2X mammalian genomes). Because these advances were published only in the latest phase of this work, we rather sought to assess the accuracy of our results post hoc. The accuracy of the 17-WA has recently been estimated through statistical inference of sequences suspiciously aligned to human chromosome 1 (Prakash and Tompa 2007). The authors estimated that BlastZ/MULTIZ algorithms performed well, with 9.7% (21 Mb) of chromosome 1 identified as suspiciously assigned. Using their data, we worked out the proportion of HCMs identified in these suspiciously aligned sequences (supplementary fig. S9, Supplementary Material online). Results ranged from 0% (chimpanzee) to 52% (tetraodon). As expected, we observed a positive trend between the proportion of microsatellites found to be “suspiciously conserved” in each species and sequence divergence, hence phylogenetic distance from human. There were less than 5% of human microsatellites in suspiciously aligned eutherian sequences, just over 10% in opossum and over 18% in nonmammalian sequences.

We chose to leave in our final data set all microsatellites found in suspicious alignments because only suspicious alignments to human chromosome 1 have been identified to date.

Discussion

Microsatellites comprise ∼3–5% of mammalian genomes (Warren et al. 2008), but little is known about their biological significance in comparison with other genomic elements, and there is still an incomplete understanding of microsatellite mutational dynamics. Despite these shortfalls and a limited success in cross-species transfer (Barbara et al. 2007), microsatellites have been widely employed as genetic markers for almost two decades. There is therefore an obvious need for comprehensive surveys of microsatellite conservation to help explore their evolution, transferability between species, possible functionality, and eventually understand their place in genomes, which will aid our general understanding of how genomes are organized.

Here, we present the first comprehensive analysis of human microsatellite conservation in vertebrate genomes. Drawing on the UCSC alignment of the human genome against the genomes of 11 mammals and five nonmammalian vertebrate species, we were able to find all human microsatellites that were conserved above the species, genus, group, or even family level. Our findings therefore significantly extend the scope of previous reports of microsatellite conservation and the sporadic identification of microsatellites conserved above the genus level in mammals (e.g., Schlötterer et al. 1991; Moore et al. 1998) and other vertebrate species (e.g., FitzSimmons et al. 1995; Rico et al. 1996).

We found that of 696,016 microsatellites identified in aligned human sequences, 85.39% were conserved in at least one other species, 28.65% in at least one nonprimate species, and 5.98% in at least three nonprimate species. On the whole, this decline of conservation appeared exponential as a function of evolutionary distance and did not necessarily always depend on time of divergence alone. Although the exponential decline of microsatellite conservation is consistent with random sequence loss (Miller et al. 2007) and thus supports the general view that most microsatellites evolve neutrally and would therefore be maintained only by chance, interpretation of such general trends is not trivial. Not only have we introduced a bias by using a human-centered alignment but the ability to tease out the dynamics of birth and death of microsatellites along the different lineages, which ultimately dictate microsatellite conservation between taxa, is missing. Should these limitations be overcome in the future with the development of a solid statistical framework, it will be possible to understand at a global scale the evolutionary trends of microsatellite retention in genomes rather than be limited to the study of individual loci.

We also found that the genomic distribution of conserved microsatellites, either at the chromosome or megabase level, was fairly homogeneous regardless of the extent of conservation, further supporting the neutral expectation that most microsatellites are maintained by chance. However, highly conserved microsatellites (NP3Ms) had a slightly different distribution (table 1), with megabase portions of the human genome containing substantially more of these microsatellites than average, providing clues that at least some microsatellites are not randomly maintained. That the decline of G + C-rich and exonic microsatellites was found to occur more slowly than that of the abundant and mutation-prone A + T-rich and nonexonic microsatellites is yet another line of evidence to support a nonneutral retention of some microsatellites in vertebrate genomes. Certainly future research should endeavor to find those microsatellites that do not follow the neutral expectation.

Overall, we believe that our method is a robust and rapid approach for identifying human microsatellites conserved in mammals, especially in eutherians, but will suffer from badly aligned sequences when applied to more distant vertebrates (supplementary fig. S9, Supplementary Material online). We recommend that these results be viewed as a preliminary attempt to characterize microsatellites conserved in nonmammalian vertebrates and, only with particular care, be used for interpretations stemming from comparisons of incomplete 2X covered genomes.

These findings raise questions as to why microsatellites might be conserved in distant species, and why microsatellites in different genomic locations are maintained to different extents. First, some regions of mammalian genomes are more “flexible,” enduring many substitutions and insertions over time, whereas other regions are more “rigid” and accumulate fewer mutations (Chiaromonte et al. 2001). Therefore, microsatellites located in constrained regions might be passively, but highly, maintained. This is concordant with our finding that highly conserved microsatellites showed both a much stronger association with other conserved genomic elements and a stronger negative relationship with SNP density than PSMs.

In addition, some microsatellite sequences may well be actively maintained. Coding microsatellites may be subject to purifying selection as they might be important for protein structure and protein–protein interactions (Hancock and Simon 2005) or to indirect selection as a source of adaptive evolution (Wren et al. 2000; Fondon and Garner 2004; Riley and Krieger 2009a). In 3′-UTRs, some microsatellites have been shown to be selected for their folding potential rather than their primary sequence (Riley et al. 2007). Although it is not clear what the function of most nonexonic microsatellites is, there is clear evidence that at least some are acting as regulators of gene expression (Kashi and King 2006; Vinces et al. 2009), suggesting that noncoding microsatellites could also be indirectly selected for mutability. Indeed, the genetic variation provided by microsatellites may be advantageous and may vary (and evolve) independently from otherwise low average nucleotide substitution rates (Kashi and King 2006). Conserved microsatellites therefore provide exciting possibilities to help single out those loci that may be actively selected for functionality, but there might be a need for further data and theoretical developments (i.e., statistical tests) to reliably distinguish between mere retention (neutral) and active conservation (selection).

Conserved microsatellites are a boon for the exploration of the mutation dynamics of microsatellites above the species level, an approach that has been rarely used to date (Zhu et al. 2000; Kelkar et al. 2008). In particular, further investigation is needed to tease out structural changes among orthologous microsatellites, for example, how compound structures arise in genomes (Kofler et al. 2008), whether there are motif changes (Riley et al. 2007; Riley and Krieger 2009b), and whether there are interspecies and intraspecies variations in length and/or mutability (Laidlaw et al. 2007; Kelkar et al. 2008). Moreover, as a consequence of the complexity and heterogeneity of microsatellite mutational dynamics, there is to date no theoretical development to estimate the life expectancy, thus the turnover, of microsatellites above the species level (Stephan and Kim 1998). We have demonstrated elsewhere that there is a strong phylogenetic signal in microsatellite loci conserved in vertebrate genomes (Buschiazzo and Gemmell 2009), therefore our data set could be fundamental for such developments and the characterization of microsatellite birth and death rates. Finally, polymorphic conserved microsatellites prove particularly useful to develop and implement transferable PCR primers (Vanpé et al. 2009). Indeed, one disadvantage of microsatellites as genetic markers is that cross-species studies needs substantial preparation (Barbara et al. 2007); provided that priming sites are also conserved between species of interest, conserved microsatellites overcome this limitation and are therefore an invaluable resource for cross-species applications in population genetics, comparative molecular ecology, and gene mapping.

Supplementary Material

Supplementary figures S1–S9 and supplementary tables S1–S4 are available at Genome Biology and Evolution online (http://www.oxfordjournals.org/our_journals/gbe/).