Our study provides the most comprehensive evaluation to date of the effect of sequence properties of microsatellites on microsatellite variability in human populations. The relatively large number of microsatellites examined here has enabled us to consider the relationships with microsatellite heterozygosity of a wide variety of sequence properties.
Our results confirm the well-known relationship between the size of the repeat unit of a microsatellite locus and the variability of the locus [73
], with larger repeat units leading to lower heterozygosity (Table ). In agreement with this trend, smaller repeat unit size was also found to lead to a higher mean number of repeats, and we observed that a higher mean number of repeats led to higher heterozygosity (Table ). For microsatellites with a single embedded STR region, loci with a di-nucleotide repeat unit had higher mean numbers of repeats (mean = 18.16) than loci with a tri-nucleotide repeat unit (mean = 13.79; P
= 2.14 × 10-12
, Wilcoxon test) and loci with a tetra-nucleotide repeat unit (mean = 12.03; P
, Wilcoxon test); loci with a tri-nucleotide repeat unit also had higher mean numbers of repeats than loci with a tetra-nucleotide repeat unit (P
= 3.33 × 10-15
, Wilcoxon test). Previous studies comparing loci with the same number of repeats but different repeat unit sizes reported the same trend that larger repeat unit size led to lower microsatellite variability [88
], suggesting that our observed relationship between repeat unit size and heterozygosity is not wholly due to the correlations of both quantities with the mean number of repeats.
We also found the composition of the repeat unit of tetra-nucleotide microsatellite loci to be an important factor in predicting heterozygosity, with repeat units high in G/C content leading to higher heterozygosity. This result agrees with a previous study [70
] that reported that of the three most common di-nucleotide repeat units in Drosophila melanogaster
(TC/AG, AT/TA, and GT/CA), microsatellite loci with repeat units GT/CA and TC/GA had higher mutation rates than loci with repeat unit AT/TA. It also agrees with the observations of a comparative genomics study of three unrelated chicken individuals [92
] that reported that tri-nucleotide repeat units high in G/C content had higher variability than tri-nucleotide repeat units low in G/C content. However, it is important to note that our results might be specific to the particular motifs available in our data set. We have only one motif that contains no G/C nucleotides (AAAT/TTTA) and only two motifs that contain two G/C nucleotides (GATG/CTAC and AAGG/TTCC), and together these motifs represent only ~1/6 of the tetra-nucleotide loci we examined (30 loci have no G/C nucleotides in their repeat motif and 23 loci have two G/C nucleotides in their repeat motif). Additionally, of the remaining 268 tetra-nucleotide loci, 253 contain the same repeat unit (ATCT/TAGA).
Our observed correlation between increases in the G/C content of the repeat unit of tetra-nucleotide microsatellite loci and increases in heterozygosity disagrees with a comparative genomics study that found that tetra-nucleotide repeat units high in G/C content led to lower variability in chickens [92
]. It also disagrees with the findings of a second comparative genomics study of human and chimpanzee orthologous tetra-nucleotide microsatellite loci that detected no significant correlation between repeat unit composition and the average squared difference in the number of repeats between orthologs [91
]. The two comparative genomics studies differ from ours in considering many more loci, but using many fewer individuals for estimating population diversity. Thus, differences in results between our study and the comparative genomics studies could arise because neither of the comparative genomics studies is entirely analogous to ours: Brandstrom and Ellegren [92
] considered data from only a small number of individuals compared to our analysis of 1,048 human individuals, and the approach taken by Kelkar et al
] is quite different from ours in being focused on genomes of different species. It is also possible that a difference arose from ascertainment of highly polymorphic loci in the genotyping panels used in our study compared to the relatively bias-free approach offered by comparative genomics. However, we have no reason to suspect that a marker ascertainment procedure selecting for variability would have produced a systematic difference in variability between different motifs. It is also possible that loci in our study might have experienced a greater degree of natural selection compared to the genome as a whole. However, a previous report by Kayser et al
] on 332 microsatellite loci with considerable overlap with the loci in our study found that natural selection did influence the vast majority of the loci. Investigating scores of the iHS test for natural selection, calculated from SNP genotype data in the three Phase I and II HapMap populations [94
] in 100-Kb regions centered on each microsatellite locus we consider here, we find that almost all loci lie within regions that have mean iHS scores that were not considered significant by Voight et al
. (mean iHS in CEU = 0.018, minimum = -1.048, maximum = 1.270; mean iHS in YRI = 0.034, minimum = -0.996, maximum = 1.797; mean iHS in ASN = 0.022, minimum = -1.331, maximum = 1.244). Thus, natural selection is not likely to have strongly influenced our results.
Our results regarding the effect of repeat unit composition on microsatellite variability also disagree with the results of Eckert et al
], who reported that tetra-nucleotide loci with one G/C nucleotide in their repeat unit (AGAT/TCTA and AAAG/TTTC) exhibited higher mutation rates than those with two G/C nucleotides (AAGG/TTCC). However, in our data (Table ), loci with repeat unit ATCT/TAGA (referred to as AGAT/TCTA by Eckert et al
]) had significantly lower heterozygosity than loci with repeat unit AAGG/TTCC (P
= 0.003, Wilcoxon test), suggesting that the differences between our results and those of Eckert et al
] are not necessarily a consequence of differences in the sequence composition of the repeat units. Our data set was obtained by genotyping 1,048 individuals for each of the 325 tetra-nucleotide loci whereas Eckert et al
] used vector-based arrays of repeats in a human B lymphoblastoid cell line. The differences between the two studies could therefore be the result of distinct cellular environments between the two studies, as our study considers accumulations of germline mutations, whereas somatic mutations were considered by Eckert et al
]. Additionally, the DNA environments differ, as we consider genomic DNA whereas Eckert et al
] examined reporter constructs.
We found tetra-nucleotide microsatellite loci containing more separate sets of repeated motifs to have generally higher heterozygosity. This observation disagrees with two previous reports that found uninterrupted arrays of Drosophila melanogaster
di- and tri-nucleotide repeats [72
] and human di-nucleotide repeats [1
] to be more polymorphic than those that had interruptions. It also disagrees with studies of vector-based poly-GT arrays in Saccharomyces cerevisiae
] and poly-CTG arrays in a human astrocyte cell line [97
] that similarly reported that interruptions in the array of repeats led to decreased variability. An important difference between our study and some of those previously reported is our inclusion of interrupted loci whose STR regions were separated by arbitrary lengths. We also applied a different threshold when defining runs of repeats, requiring four or more repeats before we considered a run of repeats as an STR region, whereas Weber [1
], for example, required three or more repeats, and Goldstein and Clark [72
] required two. Another difference between our study and that of Goldstein and Clark [72
] is that we used the total number of repeats across all STR regions at a locus, whereas their correlations with variance considered only the number of repeats in the longest run of repeats. The differences between our study and previous studies could therefore result from differences in experimental design. It is also possible that the correlation we observed between more separate sets of repeated motifs and higher heterozygosity applies to human tetra-nucleotide loci but not to other scenarios considered by previous studies.
In agreement with a previous study [90
], PCR fragment size was found to have no correlation with microsatellite variability. This is unsurprising given that PCR primer pairs are positioned so as to optimize the amplification of the locus, and their locations do not have intrinsic biological meaning. Because the distance from embedded STR regions will vary among PCR primer pairs, PCR fragment sizes do not represent absolute numbers of repeats and therefore are not comparable in a meaningful way between different loci. When we converted PCR fragment sizes into underlying numbers of repeats, however, we did find that the mean number of repeats across individuals was positively correlated with heterozygosity. Similarly, we found the maximum number of repeats across individuals to be positively correlated with heterozygosity. Some of these observations might arise from a general correlation among the various measures of diversity (Tables S5 and S6; see Additional File 5
and Additional File 6
, respectively); they are consistent with previous reports in Drosophila melanogaster
] that found the mean and maximum number of repeats to be positively correlated with the variability of di- and tri-nucleotide microsatellite loci, and with reports in humans [31
] that found the mean number of repeats to be positively correlated with mutation rate of tetra-nucleotide loci. They also agree with studies that reported that increases in the length of the repetitive component of the sequence, measured in base pairs [84
] or number of repeats [91
], led to higher rates of mutation [84
], polymorphism [92
], and average squared differences in the number of repeats between orthologous loci [91
The correlations we have observed between heterozygosity and the size and sequence of the repeat unit and the mean and maximum number of repeats are concordant with those reported between microsatellite mutation rate and repeat unit size [73
], mutation rate and repeat unit sequence [70
], and mutation rate and microsatellite length [34
]. The most commonly proposed mutation mechanism for microsatellites is replication slippage [4
]; because of homology among microsatellite repeats, the two DNA strands might realign incorrectly after polymerase dissociation and strand separation, introducing a loop in one strand and leading to microsatellite expansion or contraction after the resumption of replication [104
]. How then can our observed correlations between the sequence properties of microsatellites and heterozygosity be explained in terms of their relationship to the mutation mechanism?
The direct relationship between heterozygosity and the number of distinct STR regions and the direct relationship between heterozygosity and measures reflecting microsatellite length (mean and maximum number of repeats) might very well reflect increases in the probability of slippage as a function of the number of repeats at which it can occur [84
]. Similarly, the inverse relationship between heterozygosity and repeat unit length might reflect the increased probability of incorrect realignment after the dissociation of two DNA strands comprised of small repeated motifs compared to those comprised of large repeated motifs. For a given microsatellite length measured in nucleotides, twice as many di-nucleotide repeat units would exist compared to tetra-nucleotide repeat units, with the number of tri-nucleotide repeat units being intermediate between those of di- and tetra-nucleotide repeat units. During strand realignment, di-nucleotide repeat units would therefore have a greater chance of mispairing than both tri- and tetra-nucleotide repeat units, because of the larger number of repeated motifs present in the disassociated DNA strands; tri-nucleotide repeat units would similarly have a greater chance of mispairing than tetra-nucleotide repeat units.
Because slippage involves the loss and reforming of hydrogen bonds [106
], the influence of the sequence composition of the (tetra-nucleotide) repeat motif on heterozygosity, in which higher G/C content led to higher heterozygosity, might be attributable to the higher number of hydrogen bonds in the double-stranded DNA offered by G/C pairs that stabilize the mispaired intermediate after DNA strand dissociation and reannealing. For example, repeat unit AAGG would form 10 hydrogen bonds (two per A/T base pair and three per G/C base pair) compared to the 8 hydrogen bonds formed by repeat unit AAAT. The two additional hydrogen bonds in mispaired AAGG intermediates compared with mispaired AAAT intermediates would be expected to provide increased stability, potentially enabling more of the mispaired AAGG intermediates than mispaired AAAT intermediates to remain paired until the resumption of strand synthesis. However, with this reasoning, we would expect that the weaker hydrogen bonds for A/T pairs would cause paired strands rich in A/T nucleotides to dissociate more frequently than paired strands rich in G/C nucleotides, providing more opportunities for A/T rich sequences to undergo slippage-induced mutations. If hydrogen bonding is an important determinant of mutability, then the observation that motifs rich in G/C nucleotides lead to higher variability suggests that the effect of G/C nucleotides in stabilizing mispaired intermediates exceeds that of A/T nucleotides in generating more opportunities for mutation. Alternatively, we note that various studies have suggested mechanisms by which certain motifs might produce more mutation than others [103
], and it is possible that our observation of an effect of G/C content on variability is an artifact of a more general effect of motif composition on variability.
In conclusion, considerations of mechanisms of microsatellite mutation suggest a view in which those microsatellite sequence properties that we have observed to influence heterozygosity do so by altering the chance that a mutation event will occur. Within this perspective, increased repeat unit size acts to reduce the chance that a mutation event occurs, thereby reducing heterozygosity; increases in the number of G/C nucleotides in the repeat unit, the number of distinct STR regions, and measures of microsatellite length (mean and maximum number of repeats) all act to increase the chance that a mutation event occurs, thereby increasing heterozygosity.