Birth-/death-causing mutations

To attribute “causal mutations” to each birth/death (i.e., mutations required for a repeat to cross the microsatellite threshold), we also utilized parsimony, i.e., we selected mutational pathways requiring the smallest number of steps. We identified births caused by (a) substitutions removing interruptions in phase with the repeat (Fig. 2A); (b) deletions of interruptions out of phase with the repeat (Fig. 2B); and (c) insertions of multiples of a repeated motif (either complete or partial) (Fig. 2C; Supplemental Fig. S1C). We also identified deaths caused by (a) substitutions leading to interruptions in phase with the repeat (Fig. 2E); (b) insertions of interruptions out of phase with the repeat (Fig. 2F); and (c) deletions of multiples of a repeated motif (Fig. 2G). For mononucleotide microsatellites, we could not differentiate between a and b mechanisms for either birth (Fig. 2D) or death (Fig. 2H). More complex scenarios (e.g., Supplemental Fig. S2) usually could be assigned to the mutational types above. Multiple mutations causing a birth/death in the same lineage or in multiple lineages (Supplemental Fig. S2B) were counted independently. Following these rules, we attributed causal mutations to the vast majority of loci (e.g., ~90% for thresholds [9,5,4,3]) (Supplemental Table S2) with a low number of mutations (on average 1.03 per birth/death). Loci for which causal mutations could not be deciphered (Supplemental Fig. 2SD; Supplemental Table S2) were excluded from further analysis.

Open in a separate window

Figure 2.

Inference of causal mutation mechanisms for microsatellite birth and death (with thresholds of 5 and 10 repeats for di- and mononucleotide microsatellites, respectively). The lineage experiencing a birth (death) is marked with a green (red) star. (A) Birth by substitution (see Mechanisms of birth/death in Supplementary Information). (B) Birth by non-motif deletion. (C) Birth by motif-insertion. (D) Births resulting from either substitutions or non-motif deletions cannot be distinguished for mononucleotides microsatellites. (E) Death by substitution. (F) Death by non-motif insertion. (G) Death by motif-deletion. (H) Deaths resulting from either substitutions or non-motif insertions cannot be distinguished for mononucleotide microsatellites.

Births were more abundant than deaths at most threshold values examined (Fig. 1; Supplemental Table S2). When the threshold was set to low repeat numbers [≤7,≤5,≤4,≤3], births and deaths along the branches of the phylogeny were proportional to the respective divergence times (Glazko and Nei 2003), suggesting that, at least at the studied loci, these events did not reach saturation (Supplemental Tables S3,S4). For instance, the number of dinucleotide births (with threshold 5) along the H, C, HC, and O branches were 218, 201, 267, and 427, respectively—proportional to the divergence times of 6, 6, 7, and 13 million years (MY).

For each microsatellite motif size and repeat number examined, we computed proportions of different types of causal mutations for births occurring along any of the H, C, O, and HC branches (Fig. 3A–D). We found that the proportion of births due to substitutions (and/or non-motif deletions for mononucleotide microsatellites) consistently decreased as the threshold value used for the analysis increased. In contrast, births due to motif insertions consistently increased with increasing threshold. A marked transition in the relative proportions of birth by substitution versus birth by motif insertion occurred at distinct repeat numbers [7–8 mono; 6–7 di; 4–5 tri; 3–4 tetra]. At all repeat numbers examined, the transition-to-transversion ratios for birth- and death-causing substitutions (Supplemental Table S5) were close to the genome-wide ratio of 2:1 (Zhang and Zhao 2005), attesting to the credibility of our inference. At the thresholds corresponding to the lowest repeat numbers, motif-containing insertions and non-motif deletions contributed to births equally, suggesting low slippage rates. The proportion of non-motif deletions remained constant over all threshold values examined, potentially indicating a mechanism independent from slippage.

Open in a separate window

Figure 3.

Proportion of various causal mutations as a function of the microsatellite threshold for (A) mono-, (B) di-, (C) tri-, and (D) tetranucleotide births; and (E) mono-, (F) di-, (G) tri-, and (H) tetranucleotide deaths. Dashed lines indicate 95% bootstrap confidence intervals that were computed for each threshold by re-sampling the genome-wide set of microsatellite loci with replacement.

Similarly to births, the proportion of deaths (Fig. 3E–H) due to motif deletions consistently increased as the threshold value increased, while that due to substitutions (and/or non-motif insertions for mononucleotide microsatellites) consistently decreased with increasing threshold values. Again, the contributions of motif deletions and non-motif insertions to deaths were comparable at low repeat numbers, while that of non-motif insertions remained approximately constant over all threshold values. However, in contrast to births, deaths were primarily caused by substitutions even at high repeat numbers. Only at the highest repeat numbers examined for mononucleotide microsatellites were motif-containing deletions the dominant mechanism for deaths. Interestingly, 53 of the 8351 human-specific births and 40 of the 8010 chimpanzee-specific births (for thresholds [9,5,4,3]) constituted microsatellite “resurrection” (Harr et al. 2000), i.e., births at loci that formerly experienced death (Supplemental Fig. S2A).

Births/deaths in interspersed repeats

Next, we examined births/deaths within Alu and L1 elements, the most numerous and recently active interspersed repeats in primates. We identified elements present at orthologous locations in (1) all five primate species studied (members of the L1PA7–17, L1PB, AluS, AluJ, and occasionally AluY subfamilies), and (2) human, chimpanzee, and orangutan, but absent from macaque and marmoset (members of the L1PA3–6 subfamilies, and additional recent AluYs). Using the thresholds of [9,5,4,3] and modifying filtering criteria (Methods), we then mapped births/deaths to these elements (Supplemental Table S1) thus restricting the analysis to events that took place after their integration. Here and below we use these thresholds because for these repeat numbers we could decipher the greatest number of causal birth-death mutations (Supplemental Table S2), and at approximately similar repeat numbers we observed a shift in the mutational behavior of repetitive sequences (Fig. 3), as described above, again suggesting the proximity of the real thresholds. We also considered “stationary microsatellites” (born prior to the human-orangutan split, and not experiencing deaths since then). Gene conversion was not the dominant factor in our data set (Supplemental Fig. S3).

Births/deaths in Alus

Alu elements had similar overall numbers of births (2409) and deaths (2390). Interestingly, they were depleted in births and deaths (Alus cover 9% of the aligned human genome but possess only 4% of births and 5% of deaths; Fig. 4A, P < 0.05, χ²-tests) but enriched in stationary microsatellites (6521 loci, 18% of the genome-wide total; Fig. 4A, P < 0.05, χ²-test). These observations held when restricting the comparison to the non-coding non-repetitive (NCNR) part of aligned human genome (Supplemental Fig. S4A).

Open in a separate window

Figure 4.

Number of microsatellite births, deaths, and stationary loci mapping to (A) all Alus and different Alu subfamilies, and (B) all L1s and different L1PA subfamilies (thresholds [9,5,4,3]). Gray cells were used to derive expected counts in χ² tests for over- or under-representation of birth/death/stationary loci in all Alus and L1s (left panels), and in different Alu and L1 subfamilies (right panels). Loci corresponding to green and red colored cells have, respectively, significant under- and over-representation (P-values provided in Supplemental Fig. S10).

First, we investigated the births, deaths, and stationary microsatellites content of the three Alu subfamilies (AluY, AluS, and AluJ, inserted 13–35, 35–55, and 50–65 MYA, respectively [Batzer and Deininger 2002]) in comparison to expectations based on the relative abundance of these subfamilies in the alignments and on the total number of microsatellites belonging to each of the three life cycle stages in all Alus. Older Alus (AluJ elements) were enriched and younger Alus (AluSs and AluYs) depleted in births and deaths (Fig. 4A, P < 0.05, χ²-test), while the distribution of stationary microsatellites reflected expectations (Fig. 4A, P benchmarked at 0.05, χ²-test).

Second, we asked how births, deaths, and stationary microsatellites were distributed along the length of Alu consensus sequence (Fig. 5A–C). Births/deaths of AT-rich microsatellites were concentrated at the 3′ poly(A) tail and the middle A-stretch. Stationary AT-rich microsatellites were located at the poly(A) tails of younger (AluY and AluS) elements but were more evenly distributed between the two A-rich regions of older (AluJ) elements. Births/deaths of GC-rich microsatellites occurred at the protomicrosatellite “5′-GGGAGGCGGAGG-3′” (positions 207–218 bp) in AluS elements (Fig. 5A); AluYs have the same protomicrosatellite, but probably need additional time to accumulate the mutations. AluJ elements exhibited few births/deaths here, as their protomicrosatellite (“AGGAGTTCGAGG”) requires more mutations to become a microsatellite.

Open in a separate window

Figure 5.

Frequencies of microsatellite births (A), deaths (B), and stationary loci (C) in different Alu subfamilies along the Alu consensus sequence divided into 20-bp bins, and “frequency per alignment count” (see Methods) of microsatellite births (D), deaths (E), and stationary loci (F) in L1PA subfamilies along the L1PA consensus sequence divided into 60-bp bins (and corrected for L1PA sequence representation in alignments). In each plot, bars are decomposed and color-coded by different Alu/L1 subfamilies and by GC richness of the corresponding microsatellites (above and below horizontal midline, respectively).

Microsatellite births/deaths in coding regions

Using thresholds [9,5,4,3], a total of 157 births, 174 deaths, and 636 stationary microsatellites were identified within coding exons of the three focal primate genomes (Supplemental Table S6). Protein-coding exons were depleted in births/deaths and enriched in stationary loci compared to either the whole aligned human genome or its NCNR portion (Supplemental Fig. S3C; P < 0.05, χ²-tests). Several deaths (37 instances) took place at regions encoding polypeptide runs, and in particular in poly-Glu (10 instances) and poly-Pro (5 instances) tracts. Most births (135) and deaths (157) were caused by substitutions; 50% of birth- and 75% of death-causing substitutions were synonymous (Supplemental Table S6). Nonsynonymous birth/death-causing substitutions usually led to conservative amino acid replacements (e.g., Glu-Asp and Ala-Val) (Supplemental Table S7). Most birth/death-causing indels were within trinucleotide microsatellites; frameshifting indels leading to births/deaths of mono-, di-, and tetranucleotide microsatellites were extremely rare (Supplemental Table S6).

Microsatellite births/deaths at disease-associated loci

We aimed to determine whether the microsatellite loci associated with human diseases were recent evolutionary acquisitions or were present in the ancestors of the primate species we examined. Among the 40 microsatellite loci known to be associated with (or to be candidate loci for) human genetic diseases (Pearson et al. 2005), we could only examine 17 loci located within single alignment blocks and surrounded by unique flanking sequences (see Methods). The majority (15 out of 17) of these loci had mature alleles (i.e., above the range of threshold values) in all primate species we considered (Supplemental Fig. S7). At the folate-sensitive fragile site FRA10A locus, we found a gradual expansion of a [GCC]_n repeat to lengths above the threshold: from two in marmoset to five in orangutan to seven and eight repeats in chimpanzee and humans, respectively. This was the only case of a microsatellite birth at a disease-associated locus in our data set. For the ATXN8 opposite strand (spino-cerebellar ataxia 8) locus (ATXN8OS), all five primates (including human) possessed just two copies of the [CTG] motif, which is below the threshold to be considered a microsatellite (however, this locus expands above the threshold in patients with the disease [Pearson et al. 2005]).

Of the 15 mature microsatellite loci identified, for eight loci (PABPN1 [OPMD], COMP, FXN [FRDA], ATXN8OS [SCA8], PPP2R2B [SCA12], HOXD13, RUNX2 [CBFA1], and FRA11B) similar allele lengths were present in all species. In contrast, seven loci (JPH3 [HDL2], ATN1 [DRPLA], MAB21L1, ATXN7, ATXN1 [SCA1], ATXN2 [SCA2], and TBP [SCA17]) displayed species-specific differences in both allele length and repeat purity (Supplemental Fig. S7). Intriguingly, the human alleles were usually the longest alleles in this set of loci. For example, the MAB21L1 allele is 19 repeat units in humans but 10 units or less in other primates, while the ATN1 (DRPLA) locus in humans is 15 units in humans but 11 units (with an interruption) in orangutan.

Birth-/death-causing mutations

We found that at thresholds with small repeat numbers the majority of births took place via interruption removal by substitution. However, the dominant mechanism of births at thresholds with high repeat numbers was insertion of the repeated motif, in line with a known positive relationship between repeat number and slippage rate (Ellegren 2004). The observed transition in the relative frequency of substitution vs. slippage mutations is in agreement with a small-scale study of disease-causing mutations (Zhu et al. 2000) and a recent human-chimpanzee analysis (Pumpernik et al. 2008). We observed similar trends in the mutational dynamics of di-, tri-, and tetranucleotide microsatellites when considering overall repeat length, as opposed to repeat number. For instance, the transition in the relative importance of substitutions vs. motif insertions for births occurs at ~10–16 bp (Fig. 3). Mononucleotides differ, as the shift in birth mechanisms occurs at a smaller length for mononucleotide microsatellites, in agreement with higher rates of slippage for these repeats (Schlötterer and Tautz 1992). The similarity among di-, tri-, and tetranucleotide repeats in mutational dynamics and the transition at a specific length might be a reflection of a functional microsatellite threshold, defined as alterations in the mutational spectrum. Thus, the increased proportion of “births” by motif insertion that we detected at repeat lengths above the transition point may actually reflect mutations within mature (early adulthood) microsatellites. This interpretation is consistent with the importance of repeat length in defining a microsatellite threshold (Kelkar et al. 2010), as well as in specifying divergence (Kelkar et al. 2008) and diversity (Payseur et al. 2011) levels for mature microsatellites.

In contrast to births, microsatellite deaths at thresholds of all repeat numbers were predominantly caused by substitutions. (Mononucleotides are an exception, as motif deletions dominated at thresholds of 10 repeats or longer.) Importantly, experimental studies have demonstrated that microsatellite interruptions result in lower mutation rates in vivo (Petes 1997; Rolfsmeier and Lahue 2000; Boyer 2008), and in decreased strand-slippage errors in vitro (Kroutil and Kunkel 1999). The differential mechanisms observed for births and deaths may be explained by differences in size of the substitution target. For microsatellite birth via interruption removal, the target is a single nucleotide, which needs to be replaced by the correct nucleotide, while for microsatellite death via interruption, the target is the whole repeat, with no restriction on the substituting nucleotide. These results corroborate the few previously studied cases of microsatellite deaths (e.g., Taylor et al. 1999), wherein nucleotide substitutions played a key role in repeat degradation.

The inference of individual indel events in distant evolutionary lineages using parsimony is potentially complicated by high indel rates that may increase the chances of homoplasy. However, the effect of homoplasy on our inferences concerning mutational pathways that lead to microsatellite births/deaths is unlikely to be substantial because (1) most of the loci we considered are relatively short (Estoup et al. 2002); (2) we did not observe a saturation in the number of microsatellite births/deaths, again arguing against sizeable homoplasy; and (3) for most births/deaths, a homoplastic scenario would require identical substitutions or non-motif-containing indels (Fig. 3) occurring simultaneously at orthologous loci in multiple lineages, which is highly unlikely given the low rates of these mutations.

Transposable elements' contribution

The previously proposed importance of TEs in microsatellite genesis (Arcot et al. 1995; Nadir et al. 1996; Batzer and Deininger 2002; Buschiazzo and Gemmell 2006) was evaluated here on a genome-wide basis for the two most copious and recently active TE groups—Alus and L1s. Microsatellites can be born in TEs either concurrent to or following their integration. While we focused on the latter, our analysis of stationary microsatellites also provides information about the former.

Indeed, our results indicate that TEs play a pivotal role in the microsatellite life cycle. Remarkably, 36.3% of microsatellites (with thresholds [9,5,4,3]) present at orthologous positions in human, chimpanzee, and orangutan were likely acquired due to insertion of TEs “upon arrival” (12,794 out of 35,224 loci) (Fig. 4), with approximately equal contribution of Alus and L1s (6521 and 6273 loci, respectively). Additionally, among microsatellite births and deaths occurring in the human, chimpanzee, orangutan, or human-chimpanzee common ancestor lineages, 26.3% of births (15,360 out of 58,837) (Fig. 4) and 24.3% of deaths (11,179 out of 45,819) (Fig. 4) occurred within TEs following their integration. While interrupted microsatellites were not considered in the present study, our preliminary results indicate that such microsatellites are enriched in TEs. In fact, although Alus and L1s together cover ~25% of the aligned human genome, they harbor as many as 41% of interrupted microsatellites (126,297 of the total of 293,972 identified; P < 0.05, χ²-test), implying that TEs are important players at all stages of the microsatellite life cycle.

Based on the distribution of births, deaths, and stationary loci along the length of Alu elements in younger and older subfamilies, we propose the following model: Alus give birth to AT-rich microsatellites at 3′ ends upon integration because their 3′ poly(A) tails, essential for transposition (Arcot et al. 1995; Nadir et al. 1996; Roy-Engel et al. 2002), frequently possess long, uninterrupted [A]_n stretches. As mutations accumulate in older Alus, 3′ poly(A) tail microsatellites die due to interruptions and/or deletions, while the middle A-stretch becomes the hotbed for births and, subsequently, deaths. As a result, while most stationary microsatellites in young Alus are found in the 3′ poly(A) tails, the corresponding region in older Alus has fewer microsatellites. Integrated Alus remain an important source of microsatellites [in particular at their 3′ poly(A) tails and middle A-stretches], but are depleted in births in comparison to the genome overall, likely due to their overall GC richness. Remarkably, our results indicate that in Alus microsatellite births and deaths tend to balance out, keeping the overall number of microsatellites constant over time.

A different model emerges for L1s. They frequently give birth to microsatellites upon and after their integration; however, unlike Alus, L1s do so evenly throughout their length, potentially because their whole sequence [and not just their 3′-poly(A) tails (Boeke 1997)] is AT-rich (Szak et al. 2002). Relatedly, compared with the genome overall, L1s are enriched in births. Since L1s are on average older than Alus, they may have experienced a substantially greater number of substitutions and indels, generating more protomicrosatellites. Interestingly, we have some evidence that microsatellites in L1PAs are being eroded, likely by an excess of deaths over births; older L1PA subfamilies are depleted in stationary microsatellites and enriched in deaths.

The regional genomic context of births/deaths

The role of TEs in the microsatellite life cycle is further confirmed by our regressions. Consistent with their AT richness (Szak et al. 2002), L1s show a positive association with births/deaths, but only at 1- and 10-kb windows, likely due to their size (average ~1 kb, with some reaching 6–7 kb [Jurka et al. 2005]). Alus have a negative association with births/deaths, congruous with their GC richness. Since composition is proxied by other predictors in our fits, the observed contributions of L1s and Alus may capture additional effects. For instance, TEs are hotspots for non-allelic homologous recombination (Roy-Engel et al. 2002), potentially affecting microsatellite birth/death rates at adjacent loci (Jakupciak and Wells 2000).

Our regressions indicate a dependence of microsatellite birth/death activity on sequence composition, as captured by GC and protomicrosatellite content. Low GC content and an abundance of protomicrosatellites facilitate births/deaths, likely because such sequences have high rates of slippage (Bacolla and Wells 2004) and substitution (Arndt et al. 2005), and in agreement with a recent analysis of microsatellite abundance in protist genomes (Tian et al. 2011). We also observe a positive relationship between births/deaths and substitution rate. High local substitution rate facilitates “tinkering” with AT-rich and/or protomicrosatellites; this might result in microsatellite births, and increase the chances of interruptions in existing microsatellites, leading to their deaths (Santibanez-Koref et al. 2001).

Notwithstanding these results, the low explanatory power of our regressions suggests a limited dependence of births/deaths on regional genomic features, echoing a previous finding for mature microsatellites (Kelkar et al. 2008). However, birth/death activity might associate with other genomic features (e.g., differential efficiency of the repair), which need to be considered in future studies.

Births/deaths in coding regions and at disease-associated loci

We observed an overall deficiency of births/deaths in protein-coding exons, suggesting selective resistance to such changes within these evolutionarily constrained regions. Births/deaths by motif indels, which may involve frameshift mutations at mono-, di-, and tetranucleotide microsatellites, were particularly rare. In line with previous observations (Li et al. 2004), we found a significant enrichment of stationary microsatellites in coding exons compared with the rest of the genome. Thus, microsatellites play a critical role in encoding proteins, particularly their amino acid repeats. Note that most such repeats are encoded by interrupted microsatellites (Mularoni et al. 2010), which were not studied here.

Interestingly, the majority of the disease-associated microsatellite loci we analyzed, many of which lie within or in close proximity of coding regions, were found to be mature microsatellites (above the threshold length) in all primate species examined. This suggests that such loci are ancient, i.e., were born in a common ancestor of simians. For seven of the disease-associated microsatellite loci we examined, the human allele was longer and/or more pure than the alleles present in other primates. Both of these features, length and purity, directly affect the likelihood that the microsatellites will mutate. Thus, alleles such as ATN1 and MAB21L1 in the human genome are poised to increase in size toward a pathologic length in subsequent generations. Other loci, such as ATXN1 and ATXN2, are longer in humans than other primates, but are present as interrupted alleles in the reference genome. Such interruptions have been shown to decrease the propensity for such alleles to expand (Pearson et al. 1998). However, significant inter-individual variations in the patterns of allele interruption have been described for these loci (Sobczak and Krzyzosiak 2004), suggesting that some human populations may be more predisposed toward allele expansion to pathologic lengths. The future identification of additional microsatellite loci that are longer (and therefore more mutable) in humans than other primate species may assist in the association of new microsatellite loci with disease states.

The time scale of the microsatellite life cycle

The analyses presented here also shed light on the time scale of the microsatellite life cycle. The striking enrichment in the frequency of microsatellite deaths at the central A-rich region of the ~55-MY old AluJ vs. the ~35-MY old AluS subfamily elements indicates that the cycles of microsatellite births and deaths take place over tens of millions of years, although many longer microsatellites may persist for hundreds of millions of years (Buschiazzo and Gemmell 2010). Once born, microsatellites might be sustained for long evolutionary times. However, we observed only a few instances of microsatellite rebirth—suggesting that, once degraded, it might be difficult for sequences to resurrect from their hypo-mutating phase into dynamic microsatellites.

From the number of microsatellite births and deaths taking place along the four branches of interest (Fig. 1), we computed the average birth and death rates of microsatellites to be 1.4 × 10⁻¹¹ and 1.1 × 10⁻¹¹ per base per generation, respectively (taking into account the total length of the aligned portion of the human genome, and assuming a generation time of 20 yr). In agreement with a previous prediction (Buschiazzo and Gemmell 2006), microsatellite birth rates are higher than death rates, leading to the enrichment of microsatellites observed in primate genomes. Also, the rates of microsatellite births and deaths are lower than primate substitution or non-microsatellite indel rates (on the order of 10⁻⁸ and 10⁻⁹ mutations per locus per generation [Nachman and Crowell 2000]). This is not surprising, as de novo births/deaths (apart from births in newly integrating TEs) are restricted only to the existing protomicrosatellite/microsatellite sequences in the genome. Birth and death rates are also several orders of magnitude lower than the rates of indel mutations at mature microsatellites (10⁻² to 10⁻⁵ mutations per locus per generation [Weber and Wong 1993; Boyer and Farber 1998; Kayser et al. 2000; Ellegren 2004]). Thus, microsatellite births and deaths are indeed long-term alterations in the evolutionary fates of repetitive sequences. Note also that, due to our stringent filtering, the microsatellite birth and death rates reported here are likely to be underestimates.

Population variation at microsatellite loci

To test whether repeat numbers gathered from the reference sequences represented the variation within the species, we investigated human intra-specific sequence variation at repeat loci, using information obtained from the HapMap-ENCODE re-sequencing project (International HapMap Consortium 2003). Here, 48 unrelated individuals from three different populations—Yorubans (YRI), Europeans (CEU), and Eastern Asians (CHB+JPT)—were re-sequenced at ten 0.5-Mb ENCODE regions. In these ENCODE regions, 165 mono- and 16 dinucleotide repeat loci were found to be stationary or to harbor microsatellite births or deaths in the present analysis. At each of these loci, we identified the modal allele (i.e., the most frequent allele among the re-sequenced individuals) following the protocols in Kelkar et al. (2010) and compared it with the allele present in the reference human sequence (hg18).

Identification of births/deaths and mutations causing them

Using microsatellite presence/absence, their births/deaths and causal mutations along the H, C, HC, and O branches were inferred by parsimony (Fig. 2). For loci at which the presence/absence of information indicated a birth or death event, inference of the mutational pathway was carried out in the following four steps. (1) At each microsatellite locus, we identified nucleotides not native to the microsatellite sequence and alignment gaps in all the species. (2) Using sequence alignments at these positions, we inferred single nucleotide substitutions, insertions, and deletions sequentially from the outer nodes of the phylogenetic tree to the deeper nodes, by the principle of maximum parsimony. This takes into account the phylogenetic position of the events and uses information from the respective outgroup species (macaque and marmoset for the HC–O node and orangutan, macaque, and marmoset for the H–C node; we only considered loci at which both macaque and marmoset either possessed or did not possess a microsatellite). We then used the parsimony-derived ancestral sequences thus inferred at evolutionarily recent nodes for inference on mutations along deeper lineages within the tree. (3) Among all mutations identified, we only considered those that allowed a sequence to cross a particular threshold. For instance, while considering microsatellite birth in Figure 2A, we focused on the interruption removal in the HC lineage, while ignoring repeat number differences between human and chimpanzee. (4) We placed each mutation event into one of the classes illustrated in Figure 2. At each threshold, we obtained the 95% confidence intervals for the observed genome-wide proportions of all these mutation classes by generating 1000 bootstrap samples (re-sampling with replacement) of the genome-wide set of microsatellite loci, and computing the 2.5th and 97.5th percentile proportion values.

Microsatellite births inferred to have taken place by substitution of interrupting nucleotides (for instance, Fig. 2A) may also be caused by a slippage-related deletion involving the interrupting nucleotide and, sometimes, flanking bases (in Fig. 2A, the deletion of “CT”). However, given that slippage-related indels are rare in interrupted microsatellites (Brinkmann et al. 1998; Boyer et al. 2002; Sibly et al. 2003), and that slippage-related insertions were found to be rare at small repeat numbers in our own data (Fig. 3), we concluded that the most likely mechanism involved in interruption removal was substitution.

For the majority of microsatellites, the inferred mutations agreed with microsatellite birth/death inferences based on presence/absence (Supplemental Table S2); in a number of cases (e.g., ~10% for thresholds [≤9, ≤5, ≤4, ≤3]), where the presence/absence-based inference was conflicting with the mutational pathway inference, the birth/death scenario revealed by the mutational pathway was given priority. All such cases involved multiple mutations in the same locus. For instance, if at a particular locus marmoset, rhesus and orangutan had an [AC]_i microsatellite with repeat numbers above the threshold (e.g., five repeats), and at the orthologous position, the chimpanzee non-microsatellite sequence was [AC]₄TC[AC]₂ and the human sequence was [AC]₄TC[AC]₇, then based on microsatellite presence/absence we would infer a death along the chimpanzee lineage. However, the mutational pathway analysis revealed the existence of two separate birth/death events: an A→T substitution leading to a death in the human-chimpanzee lineage, followed by motif insertions leading to a birth in the human lineage. For loci with multiple birth/deaths along different lineages (0.1% of all loci), each event at a locus was counted independently.

Births/deaths in transposable elements (TEs)

RepeatMasker (http://www.repeatmasker.org/) hg18 SINE, LINE, and coding region (refseq) annotations were obtained from the UCSC Genome Browser (http://genome.ucsc.edu/). Consensus sequences for TE subfamilies were obtained from RepBase (Jurka et al. 2005). Portions of the genome that did not overlap with TEs and coding exons were classified as non-coding non-repetitive (NCNR). Microsatellites whose coordinates intersected with those of the TEs were identified as TE microsatellites loci. The position of each TE microsatellite relative to the consensus sequence of its enclosing TE's subfamily (i.e., the consensus-relative position) was determined. For microsatellite loci within Alus, their position relative to the universal Alu consensus sequence (Fig. 5A) was determined by first aligning (using ClustalW with default options) the enclosing Alu's subfamily consensus sequence to the universal primate Alu consensus sequence obtained from RepBase (Jurka et al. 2005). A similar procedure was implemented for microsatellites identified within L1PA elements. Since the focus here is on the role of Alu's and L1's in microsatellite births/deaths (and not on mutational pathways leading to them), we relaxed our filtering criteria by allowing inter-microsatellite distances >10 bp, and permitting lower sequence identity (down to 80%) of flanking sequences between aligning species (Supplemental Table S1). χ²-tests for motif abundance in TEs of different ages were implemented in R (http://www.r-project.org/). P-values were Bonferroni-corrected. To account for the 3′ biased representation of L1s (Boissinot et al. 2001), birth/death frequencies in 60-bp bins of the consensus L1 sequence were divided by the number of times each such bin occurred in the alignments (Fig. 5B).

Evolutionary history of disease-causing microsatellite loci

We used Galaxy (http://galaxy.psu.edu) to obtain alignments of 40 disease-associated loci (and their 10 bp upstream and 10 bp downstream flanking sequences) using their previously published hg18 coordinates (Pearson et al. 2005). We identified 17 loci that (a) were present in single alignment blocks, (b) were flanked by at least 10 bp of non-microsatellite sequences upstream as well as downstream, and (c) were represented in at least three of the hg18, panTro2, ponAbe2, rheMac2, and calJac1 genomes. The other 23 loci were excluded primarily because they contained very long microsatellites broken between alignment blocks and thus we could not infer their lengths accurately.

Multiple logistic regression analysis

Multiple logistic regressions were conducted in R (http://www.r-project.org/) to contrast births/deaths (event) vs. control windows. For every window size considered, event windows were positioned to maximize the number of events per window (e.g., if two events were <1 kb apart, and the window size used was 1 kb, both events were included into a single event window of 1 kb in size). Control windows were placed randomly, but at least 100 bp away from the boundary of any event window, and were equal in number to the event windows. For each window size and each window, we computed the following predictors: recombination rate, recombination hotspot density, coding region content, CpG islands content, SINE content, LINE content, distance from a telomere, human-orangutan substitution and indel rates, and the fraction of the window covered by protomicrosatellites. Human fine-scale regional recombination rates and hotspot locations (Myers et al. 2005) were used to compute, for each window, average recombination rate, and hotspot density. The hg18 annotations for coding regions, CpG islands, SINEs, and LINEs were obtained from the UCSC Genome Browser and used to compute the corresponding contents per window. Distance from telomeres was computed using the UCSC Genome Browser (Fujita et al. 2010) annotations for hg18. Human-orangutan substitution and indel rates were chosen over human-chimpanzee substitution rates to minimize ancestral polymorphisms' influences on these estimates. An additional predictor calculated in each window was the fraction of the window covered by proto-microsatellite sequences (stretches of sequence at least 10 bp in length, of which at least eight are occupied by the same nucleotide; such sequences may give birth to the most common microsatellites). For each window, the number of nucleotides (used to calculate predictor densities and contents) was measured by subtracting from the window size (either 0.1 Kb, 1 Kb, 10 Kb, 50 Kb or 100 Kb) the number of hg18 nucleotides not covered by alignments, and the number of aligned hg18 nucleotides covered by microsatellite loci that were filtered out during our orthologous microsatellite search (see Results).

Filtering steps were implemented before running the regressions. In particular, only event and control windows with high alignment coverage (>95% of window length) were used. For each regression, we also discarded extreme outliers with high influence (Cook's distance of 1 or greater). After filtering, we selected a satisfactory logistic regression by sequentially eliminating predictors based on their individual contributions to model performance, Akaike Information Criterion scores (with/without each predictor) and predictors' variance inflation factors (which were required to be <10 [Kutner 2005]). The performance of a logistic regression is measured as the share of deviance explained, (D_o – D_m)/D_o, where D_o indicates the null deviance, and D_m the residual deviance carried by the model. Accordingly, the individual contribution of a predictor within a model is measured as ([D_o – D_m] − [D_o – D_m{−}])/(D_o–D_m), where D_m and D_m{−} are, respectively, the residual deviances carried by the model and by the reduced model obtained removing the predictor in question.

We thank Astrid Roy-Engel and three anonymous reviewers for important suggestions, and Guruprasad Ananda for assistance with Galaxy. This project was supported by NIH grant R01-GM087472.