Repeat-encoded poly-Q tracts show statistical commonalities across species

Cross-species occurrence of tri-nucleotide repeats

As a first step towards understanding species-specific characteristics of TNRs, we identify and analyse TNR tracts in six different species: Saccharomyces cerevisiae, Arabidopsis thaliana, Caenorhabditis elegans, Drosophila melanogaster, Mus musculus and Homo sapiens. Repeat tract scanning identified 247, 1947, 559, 3996, 79727 and 35736 TNR sequences in these species, respectively. As repeat length and repeat sequence are widely accepted factors in TNR variability, we compare these properties across organisms.

Length distribution of triplets was broadly similar across species (see Additional file 1: Figure S1). However, comparison of the distribution of repeat unit sequences showed more interesting patterns (see Figure 1). In all organisms, distribution of sequences in identified TNR tracts was significantly different to the background triplet frequency (p≪1e⁻¹⁴ for all species, using a chi-squared test against a genome-wide order-two Markov background), agreeing with earlier genome-wide analyses of human repeats [12]. More interestingly, the distribution of triplet sequences in different organisms demonstrated markedly different patterns of non-randomness, even after compensating for different backgrounds (i.e., using per-organism and per-chromosome backgrounds, comparing log-ratios across organisms).

It is interesting to note the large differences in TNR frequencies among these genomes. Notably, the Drosophila melanogaster genome (∼165Mb) contains more than six times as many TNR sequences as the Caenorhabditis elegans genome (∼100Mb); there are over twice as many TNRs in the mouse genome than in the human genome. The latter is particularly striking since their genomes are similar in size and the large degree of homology between them.

These analyses suggest that simple uses of known correlates of TNR expansion are unlikely to produce informative cross-species patterns. While the differing distribution of triplet sequences may be curious, it does not provide any new insights into the structural characteristics or function of TNR sequences. As an alternative approach, we focus on higher-level characteristics such as the known functional associations and the homo-AA tract composition of TNR sequences.

Cross-species functionality of tri-nucleotide repeat sequences

Human gene-associated repeats – particularly coding region repeats – are known to have overrepresented GO terms indicative of regulatory function [12]. To test whether these functional associations are a cross-species phenomenon, we performed an analysis on the mentioned species. We isolated TNR sequences in these species’ genomes that could be localised to coding regions according to genomic feature annotations (See Figure 2). This set of repeat-containing genes were then analysed for systematic over-representation of GO terms. We found that, in several species, TNR-containing coding sequences showed over-represented regulation-associated GO terms (see Table 1), supporting a view that these associations are a cross-species phenomenon.

From these results we conclude that the previously-observed regulation association of coding region TNRs is not exclusively a human-specific trait, but can be seen as a cross-species phenomenon, even across a range of dissimilar organisms. Importantly, these results do not address the possibility that the functional associations are the result of a derivative sequence property, such as the homo-amino acid repeat tracts in corresponding proteins.

Homo-amino acid repeat sequences and tri-nucleotide repeats

Previous GO over-representation analyses of TNR se- quences have used whole organism gene sets as a statistical background [12]. However, when looking at functional associations of genes containing coding sequence-localised TNR tracts, it must be noted that the protein sequences associated with these genes will constitute an unusual subset of the proteome, and may give a very different statistical background. Specifically, translations of TNR tracts will result in protein sequences enriched in homo-AA tracts.

Through their association with TNRs, protein homo-AA repeat tracts have been implicated in a range of human diseases [16] and are more likely to be involved in transcriptional regulation [15], possibly due to the structural characteristics of the homo-AA tract. It has been suggested that these tracts are inherently structurally disordered [17–19], and that such unstructured regions may act as flexible regions, increasing binding affinity [18]. The prevalence of transcription factors amongst homo-AA repeat-containing proteins raises the question of whether functional associations previously ascribed to coding sequence TNRs may be explained by the homo-AA tracts they encode.

Due to the redundancy of the genetic code, a homo-AA tract is not necessarily encoded by a TNR tract; instead, variant encodings may be used and in fact, they are expected to be less prone to mutation-driven length variability. As such, repetitive DNA sequence encoding important regulatory functional elements appears less than optimal, unless there is a associated functional difference.

In order to identify the degree to which homo-AA tracts in TNR-associated proteins explain the functional associations of these nucleotide repeats, we performed whole-proteome repeat scans of each organism’s non-redundant proteome and split the identified homo-AA tract containing proteins (hereafter simply referred to as homo-AA proteins) into two sets – those where the homo-AA tract was encoded by a repetitive DNA sequence, and those where variant encoding was in use – before identifying functional associations of these sets using GO terms.

GO term over-representation testing of the TNR-encoded homo-AA proteins was initially done using the variant-encoded homo-AA protein set as a background model; this test identifies whether the TNR-encoded set is significantly different from the variant-encoded set. We also used the whole set of homo-AA containing proteins as a background, identifying whether TNR-encoded proteins form an identifiably distinct subset of all homo-AA proteins in terms of functional associations.

Homo-amino acid tracts and sequence stability

Increased sequence instability is a distinguishing feature of TNR tracts as a whole; in TNR-encoded protein-coding repeat regions, such instability underlies the repeat diseases, but may also affect other aspects of protein function, such as the number and type of interactions the encoded protein is involved in. However, similar instability would not be expected in variant-encoded tracts. In order to identify whether sequence instability is a distinguishing factor of TNR-encoded homo-AA tract proteins, we investigated two characteristics related to sequence stability: protein-protein interaction (PPI) counts and estimated sequence conservation.

A protein’s number of PPIs and its evolutionary rate have been shown to be linked; it has been observed that proteins with higher PPIs evolve more slowly, likely due to sequence constraints involved in maintaining existing interactions [20], though other factors such as expression levels also contribute [21]. As such, the higher variability commonly associated with TNR tracts suggests that homo-AA proteins should have lower PPI counts than their variant-encoded counterparts.

Using the same TNR- vs. variant-encoded distinction as above, and PPI data from the IntAct database, we identified the number of PPIs each homo-AA protein is involved in. Looking at the distribution of PPI counts in these proteins (see Additional file 2: Figure S2) we find that there is no significant difference between TNR- and variant-encoded homo-AA proteins in terms of the number of protein interactions associated with each set.

A different approach to investigating sequence stability is to directly assess the conservation of the homo-AA encoding sequence itself. We used pre-computed Drosophila and human PhastCons scores from UCSC (see Methods) to evaluate sequence-level conservation. Genomic loci corresponding to homo-AA encoding regions were obtained by reverse-mapping homo-AA tract boundaries onto exonic sequence. From these loci and PhastCons scores, we obtained conservation metrics for individual tracts. Segmenting these conservation scores as above, we found that conservation of homo-AA encoding DNA was not significantly affected by whether the sequence was classified as a TNR.

Note that this finding does not contradict previous findings that TNR sequences show higher variability. The comparison here is with variant-encoded homo-AA sequences, which constitute a very specific background model. In addition, PhastCons scores are not well-suited to identifying repeat length variation; as such, this method will not account for a major factor in the variability of TNR tracts.

Homo-amino acid tract composition

As repeat unit and repeat length are central factors in determining TNR variability, considering these factors is also essential when investigating homo-AA proteins. Using the same TNR- and variant-encoded protein sets as above, we classified homo-AA proteins by the repeated residue and compared the frequency and length of residues between the sets. The hypothesis was that there would be no difference between the TNR- and variant-encoded sets in terms of amino acid make-up of repeat regions.

Looking at residue frequencies, we found that for human, mouse and fly, glutamine repeats were significantly more likely to be encoded by TNRs than by variant encoding while proline was significantly less likely to be TNR-encoded (see Figure 3). TNR-encoded tracts also tend to be more prevalent to code for glutamic acid and asparagine repeats in most species. In terms of length, we found that in all organisms, the average TNR-encoded poly-glutamine repeat tracts were longer than in their variant-encoded counterparts; the results were significant for human, mouse and fly (see Figure 3); for human only, homo-AA tracts of a number of other amino acids (alanine, aspartic acid, glutamic acid, glycine, lysine, leucine, proline and serine) were also significantly longer in TNR-encoded homo-AA proteins (see Additional file 3: Tables S1 and Additional file 4: Table S2 for more detail). These findings show that poly-glutamine tracts are notably different when encoded by TNRs, and that the differences are consistent with characteristics of human repeat disease, as discussed below.

Genomic data

The human (hg19), mouse (mm10), Drosophila melanog- aster (dm3) and Caenorhabditis elegans (ce10) reference genomes and genomic feature locations were obtained from the UCSC Genome Browser [22] (RefSeq Genes track [23]). Annotations for Arabidopsis thaliana and Saccharomyces cerevisiae were from TAIR [24] (tair9) and SGD (S288c) [25], respectively. Mitochondrial, chloroplast and unassembled chromosome sets were excluded from further analysis. Multiple splice variants were not considered; in each case, all but one splice variant was discarded. For tair9, the first identified splice variant was retained in the absence of canonical splice information.

From each annotation, we retained the largest set of genes so that there is a unique mapping between the gene identifiers and Uniprot protein identifiers. Feature locations were used to classify regions as intronic, exonic (i.e., coding region), $5^{\prime }$ or $3^{\prime }$ UTR, upstream or intergenic; these mutually exclusive classifications were then used to annotate genomic repeat tracts.

Genomic repeat tracts

Repeat tracts were identified using Tandem Repeats Finder 4.04 [26], with the following parameters: Match=2, Mis456, match=7, Delta=7, PM=80, PI=10 and Minscore=40. and a maximum repeat period of 3. Identified repeats were further filtered to remove all repeats with a period of one or two; period-one repeats have multiple periodicities, but were here considered to be mono-nucleotide repeats and were excluded from further consideration.

For comparison with other definitions of a repeat, the minimum length under this scoring is $6\frac{2}{3}$ repeat units (i.e., 20 nucleotides) with no mismatches. In subsequent tests involving amino acid tracts (see below), we used a stricter criterion to enable precise reverse mapping from amino acids coding sequence.

Sequence frequency analysis

For each genome, whole-genome and per-chromosome tri-nucleotide frequencies were determined using a custom script to obtain an order-two Markov background from the whole-genome and chromosome sequences respectively. In order to test whether TNR sequence frequencies were consistent with the (order-two Markov) background, a chi-squared test was used.

Gene Ontology (GO) associations

GO term associations with genes/gene products were obtained from the Gene Ontology project, as was the ontology itself (version from July 2012). Individual genes or gene products were annotated with each GO term appearing in the association set, and with the transitive closure of those terms. The transitive closure of the GO graph was constructed using only is_a and part_of relationships to avoid false positives (e.g., from has_part or regulates relationships). GO term over-representation was assessed using the Fisher exact test, with Bonferroni correction applied to adjust for multiple hypothesis testing. In preliminary studies, GO-term over-representation in non-coding regions was analysed. The associations discovered were weaker and semantically very similar to those of coding regions, which also have more common disease associations; as a result, non-coding regions were not included in further analyses.

Proteomic data

The proteomes were downloaded from UniProt, using the UniProt/Swiss-Prot identifiers obtained by mapping gene identifiers from the annotations described above.

Protein-protein interaction data were sourced from the IntAct database [27].

Amino acid repeat tracts

Repeat tracts were identified by scanning all protein sequences for homo amino acid runs of length at least seven residues for correspondence with the repeat unit thresholds identified by Tandem Repeats Finder. We then examined the coding sequence for each homo-AA tract to determine if it is encoded by a TNR: a homo-AA tract is considered TNR-encoded if at least seven consecutive residues of the tract were encoded by the same codon. (A separate tool XSTREAM [28] is available to identify homo amino acid runs, but due to small discrepancies of what counts as a repeat by Tandem Repeats Finder and XSTREAM we were unable to utilise them to map back and forth between genomic and proteomic repeat locations.) The complete proteome sets were scanned and repeat sequences identified were used as a base set for further study.

Genomic conservation scores

As a measure of per-site genomic conservation, pre-computed PhastCons [29] scores were used. These were sourced from the UCSC genome browser tables for Drosophila (phastCons15way) and human (phastCons46way).

We did not complete this analysis for the other four organisms.

Tract composition analysis

Analysis of tract composition was undertaken for TNR-encoded homo-AA tracts. The distribution of specific amino-acid repeats encoded by TNRs was assessed by a two-tailed binomial test for each amino acid; success counts were defined as the number of TNR-encoded repeats for that residue, with the probability of success defined as the proportion of TNR-encoded homo-AA proteins over the total set of homo-AA proteins.

Length of homo-AA repeats in TNR- and variant-encoded tracts was compared using the non-parametric Mann-Whitney U test. All homo-AA tract lengths were gathered, split into one set per residue, and annotated as being either TNR- or variant-encoded. A significant result indicates that homo-AA repeats tracts of a given amino-acid are longer (or shorter) when TNR-encoded.