We analyzed 33,860 human peptide sequences from the Ensembl database 12 for the presence of tandem amino acid repeats of size 5 or longer; 5,467 proteins contained at least one such tandem repeat (about 16%). The most common amino acid repeat types (n > 200) were glutamic acid (888), proline (883), alanine (681), serine (623), glycine (510), leucine (392), glutamine (273) and lysine (223). The average repeat size was similar for different amino acids (in the range 5.8 to 6.8) except for glutamine, with longer repeats (average 8.7).

We mapped the human ESTs 13 to the repeat regions using TBLASTN 14. We selected those repeat regions, including the tandem repeat and 15 nucleotides of adjacent sequence at each side, that were covered by at least 4 different ESTs with >90% identity matches. These comprised 2,227 repeat regions, about 41% of the initial ones, with an average repeat coverage of 27.4 ESTs per repeat. Within these, 115 (5.2%) showed one or several polymorphic variants, each supported by at least 2 different ESTs (Table 1). The amount of polymorphism varied between different amino acids, from 2.4% in leucine repeats to 10.2% in glutamine repeats. Considering only cases for which we had 100 or more ESTs, repeats that were polymorphic went up from 5.2% to 21% (26 out of 123).

We detected 137 different polymorphic variants in the 115 polymorphic amino acid repeats. We classified them as those containing gaps (or indels) of which there were 60 (43.8%), and those containing only amino acid substitutions, of which there were 77 (56.2%) (Additional data file 1) We also measured the repeat codon homogeneity as the fraction of the repeat encoded by a perfect codon run. A high average codon homogeneity in the sequences encoding different types of amino acid repeats was generally associated with a high percentage of gap polymorphic variants (Table 1). Glutamine repeats showed the highest frequency of gap polymorphisms (88% of the glutamine polymorphic variants) and the highest average codon homogeneity (0.66), while proline and serine repeats showed the lowest gap polymorphism frequencies (14% and 18%, respectively) and the lowest average codon homogeneity (0.41 in both cases).

We also analyzed polymorphisms supported by ≥4 ESTs, which should be enriched in more common polymorphic variants; they comprised 38% of the polymorphism data. In this dataset, the frequency of repeat gap polymorphisms was higher than that of substitutions (35 versus 17).

We compared the repeat polymorphism levels to those of the sequences immediately adjacent to the repeats, considering, at each side of the repeat, a sequence of the same length as the corresponding repeat (Additional data file 1). The overall number of polymorphic variants was similar to that found within the repeats (4.8% versus 5.2%), but the number of polymorphisms containing gaps was remarkably lower, 8 in upstream regions and 14 in downstream regions, about 5 times less than within repeats (60 cases). In contrast, substitutions were slightly more common outside the repeats than inside them: 103 and 93 in upstream and downstream regions, respectively, compared to 77 within repeats.

Among polymorphisms supported by ≥4 ESTs, the trend was maintained for a larger number of gap polymorphisms within repeats than outside repeats (35 in respect to 3 in upstream and 11 in downstream sequences) and only a small difference for substitutions (21 in upstream and 26 in downstream sequences, in comparison to 17 within repeats).

We compared the number of polymorphisms involving gaps or substitutions in different amino acid repeat types and adjacent regions (Figure 1). This analysis showed that the previously observed larger number of substitutions outside the repeats with respect to inside the repeats could be mainly attributed to leucine repeats (10 and 12 polymorphic variants in upstream and downstream sequences, respectively, versus only one within the repeat). On the other hand, glutamine, alanine and glutamic acid repeats were the main contributors to the increased number of gap polymorphisms inside the repeats than outside the repeats. The ratio between gap polymorphisms and substitution polymorphisms was highest in the case of glutamine (15 versus 2) and lowest for proline (3 versus 22), indicating strong differences in the susceptibility to slippage of different repeat types.

We investigated the frequency of the different amino acid substitutions in repeats and adjacent sequences, focusing on the eight most common amino acids forming tandem repeats (Table 2). The aim was to identify possible biases in the amino acid substitution patterns inside repeats with respect to the repeats' adjacent sequences, as this could be informative of specific selective constraints operating in the repetitive structures. The dataset analyzed comprised 79 substitutions inside repeats and 135 in adjacent regions. In the first place, we determined that the vast majority of amino acid substitutions could be explained by single non-synonymous nucleotide changes. Inspection of the types of amino acid substitutions in repeats and adjacent sequences indicated that there were no major differences between them. For example, nearly all amino acid substitutions that occurred at least five times in the adjacent sequences, representing the most common amino acid replacements, could also be observed inside the repeats. The only exception was the replacement of G by V, with seven cases in adjacent sequences versus none within repeats. Given the low number of cases, however, this observation should be treated with caution.

In addition, we inspected the relative position of substitutions inside the repeats. This could be informative for biases in the positions where substitutions more often occurred. For example, an excess of substitutions at the repeat extremes could indicate a selective pressure to preserve a minimum length of the repeat. However, the observed position of amino acid substitutions was overall not significantly different from the expected distribution if substitutions were located at random (see Materials and methods and Additional data file 1). In conclusion, this analysis did not detect any specific differences in the selective constraints related to different amino acid substitutions inside or outside repeats, or in the relative position of the substitutions within the repeats.

Another aspect we studied was the relative frequency of synonymous and non-synonymous substitutions inside repeats and in repeat adjacent regions. We identified all the synonymous and non-synonymous nucleotide changes in the EST dataset and divided it by the total number of synonymous and non-synonymous positions analyzed (Figure 2 and Additional data file 1). In the two types of regions, the frequency of non-synonymous substitutions was lower than that of synonymous substitutions, as expected if some of the substitutions resulting in amino acid changes were negatively selected. The frequency of synonymous substitutions was very similar inside and outside the repeats: 0.015 (1.5% of sites) inside repeats and 0.014 to 0.016 (1.4% to 1.6% of sites) in repeat upstream and downstream regions, respectively. In agreement with the results obtained on amino acid substitutions, the frequency of non-synonymous substitutions was similar inside repeats (0.009, 0.9% of sites) and outside repeats (0.011; 1.1% of sites, in both repeat upstream and downstream regions). By amino acid type, only proline and glutamine repeats showed a non-synonymous substitution pattern different from their corresponding adjacent sequences. In the case of proline, the frequency of non-synonymous substitutions was 0.02 while the average of the two adjacent regions was 0.012. That is, there appeared to be an almost two-fold excess of non-synonymous changes within repeats. In the case of glutamine repeats, the opposite trend was observed, with a non-synonymous substitution frequency of 0.005 inside the repeats versus 0.011 in the adjacent regions.

We next compared the codon homogeneity values of all repeats to those of the repeats associated with gap polymorphisms or with substitution polymorphisms (Figure 3). The average value was 0.49 in all the repeats analyzed, 0.44 for those with substitution variants and 0.65 for those with gap variants. Repeats that showed gap polymorphisms had higher codon homogeneity values than average (p = 0.001, Kolmogorov-Smirnov test). Those with substitution variants, instead, were similar to the general repeat population. These results are expected if we consider that slippage will mainly act on long pure codon tracts, resulting in expansions and contractions of the repeats. Interestingly, however, the presence of a long pure codon tract does not appear to be indispensable for this type of polymorphism to occur, as in about 25% of the cases the longest pure codon run had a short size, between 1 and 3 codon repeat units.

Polymorphic cases related to the expansion or contraction of repeats, those that involve gaps, are of particular interest because of the potential of these elements to cause disease. Table 3 lists genes containing this type of polymorphism for the most abundant amino acid repeat types. Among them we detected two poly-glutamine containing genes known to be associated with neurodegenerative disorders: dentatorubral-pallidoluysian atrophy protein (DRPLA) and spinocerebellar ataxia protein 6 (voltage-dependent P/Q-type calcium channel alpha-1A subunit, or CACNA1A). These two disease loci contained long runs (19 and 13 glutamines, respectively) and high codon homogeneity levels (0.79 and 1, respectively). Other genes in the list with long homopeptide runs and high codon homogeneity are thus possible candidates to be associated with disease. Among genes showing expansion/contraction polymorphisms was an abundance of transcription factors and RNA-binding proteins. Most of the polymorphic variants were one repeat unit away from the reference repeat, the maximum difference being three repeat units. The detection of longer repeat size variants would be hindered by our >90% identity EST match criteria, but, given that only two variants were found that show a 3 repeat unit size difference, these cases are expected to be rare. In 12 cases, the longest pure codon run occupied the totality of the repeat (codon homogeneity of 1). Length polymorphisms were most frequently associated with CAG (glutamine), GAG (glutamic acid) and CTG/GCT (leucine/alanine).

Human protein and cDNA sequences were extracted from the Ensembl database (NCBI35-based release) 12. The number of initial peptide sequences was 33,860. The source of EST sequences was the NCBI-EST database (Feb 3 2005) at the National Center for Biotechnology Information 13, containing 5,430,499 EST human sequences.

We used our own programs to identify all single amino acid tandem repeats of size five or longer in the human proteins and to extract the DNA sequences encoding them. We identified repeats in 5,467 different proteins. For each repeat we stored the repeated amino acid, position in the sequence, repeat length, length of the longest pure codon run and codon(s) in the longest pure codon run(s). In specific cases we also retrieved the equivalent repeat in the mouse and rat orthologous sequences using BLASTP 14 at NCBI. For each repeat we calculated codon homogeneity as the fraction of the repeat occupied by the longest pure codon run. The non-parametric Kolmogorov-Smirnov test was used to assess the difference in the codon homogeneity values of different samples.

We mapped all human ESTs to the repeat regions in the reference proteins using the program TBLASTN 14. The repeat regions included the perfect tandem repeat and 15 nucleotide sequences at each side of the repeat. We considered only EST matches that covered the entire repeat region and showed a percent identity >90%. This may hinder the detection of very divergent polymorphic variants but limits the chances of matches between unrelated sequences. For analysis we selected those repeat regions that were covered by at least 4 different ESTs (2,227 cases). We also retrieved cases covered by at least 100 different ESTs (123 cases).

Polymorphic variants were identified as changes to the original sequence supported by at least 2 independent ESTs (137 within repeats, 111 in upstream regions, 107 in downstream regions). We counted the different types of polymorphisms within the tandem repeats and in sequences of the same length immediately adjacent to the repeats. We discarded those cases where adjacent regions also contained repeats. For comparison we also analyzed polymorphic variants supported by at least 4 ESTs (52 within repeats, 24 in upstream regions, 37 in downstream regions). They were classified as variants involving expansions and/or contractions (gaps or indels) and variants involving only amino acid substitutions.

We counted the observed frequency of all possible types of amino acid substitutions, in the repeat and adjacent sequence polymorphic variants, for those amino acid repeat types that were most frequently found in tandem repeats (A, E, G, L, P, S, K, Q). No strong differences were observed in the two datasets. We also counted the position of substitutions within the repeats, by assigning each substitution to one of the following classes: pos = 1 (first position of the repeat), pos = 2 (second position), pos = -1 (last position), pos = -2 (position before the last one) and middle (remainder of positions). We calculated the expected values under a random distribution; for example, in a repeat of size 5, each class will have an expected value of 0.2, and in a repeat of size 6 all classes will have an expected value of 0.16 except the middle, which will have an expected value of 0.33. The total expected values for each group were compared with the observed values using a chi-square test. No significant differences were found.

We counted the observed number of synonymous and non-synonymous nucleotide substitutions in the non-redundant EST dataset matching the repeats and their adjacent regions. In this case we included substitutions represented by a single EST as well as by several identical ESTs to have a sufficiently large dataset to be able to obtain and compare substitution frequencies. Some of the changes could be due to sequencing errors. This type of error should affect both synonymous and non-synonymous substitution rates inside and outside the repeats in the same manner. As we still detected differences between the two types of rates, and as our main goal was to compare different regions and types of homopeptides, we used all the observed mutations in the non-redundant EST dataset. To maximize the reliability of the alignments we discarded ESTs containing gaps (3%). The dataset comprised 8,196 different non-redundant ESTs. We counted the number of synonymous and non-synonymous positions analyzed to obtain the frequency of substitutions of each kind. Overall, we analyzed 430,161 nucleotide positions, 107,845 of which were synonymous and 322,316 non-synonymous. The total number of synonymous substitutions was 1,663 (1.54% of sites) and of non-synonymous substitutions 3,458 (1.07% of sites). We also extracted the results for sequences containing each different amino acid repeat type.