Tandem repeats discovery service (TReaDS) applied to finding novel cis-acting factors in repeat expansion diseases

Overview on repeat expansion diseases

At present 29 diseases are classified as repeat expansion diseases (RE) [1–3], and the number is growing. These are mostly neurodegenerative and neuromuscolar disorders, including Huntington disease (HD), Kennedy disease (SBMA), and several types of Spinocerebral Ataxias (SCA). Since up to recently all known cases involved repeating a motif of 3 nucleotides, this class was denoted also as trinucleotide repeat (TNR) expansion disease. However, cases of repeating units with 4, 5 and 12 nucleotides have been discovered thus we talk more generally of repeat expansion diseases. Recent surveys devoted to DNA repeats [4] have extended discussion of repeat expansion disorders, while specific surveys for repeat expansion diseases can be found in [5–7].

The locus of expansion can be located in various regions of the resident gene: in the coding sequences, in the 5'- untranslated region (5'-UTR), in the 3'- untranslated region (3'-UTR), in introns and in promoter regions. Two main questions are related to the study of these diseases from a genetic point of view: (a) which mechanisms or conditions lead to the repeat expansion? and (b) how do repeat expansions result in diseases?

Only a small fraction of all the tandem repeats found in the human genome expand and result in a disease. Thus researchers have tried to identify which unusual structural features favor such expansion, and found a propensity to forming hairpins (or other structures, such as: quadruplex-like structures, H-DNA and sticky DNA) as a key mechanism leading to expansion. Several studies also tried to identify cis-regulating elements that do favor the onset of the above structural features and of the expansion. Our study falls in this category and proposes long fuzzy tandem repeats as a novel cis-regulating element for repeat expansion, thus contributing to investigating question (a).

Cis-acting factors for TNR instability

Several papers tackle the problem of determining cis-acting factors associated with loci of TNR instability. In particular one quite studied factor is the proximity and orientation of DNA replication initiation regions (IR) w.r.t. the TNR instability locus [8, 9]. In [8] the position of the DNA replication initiation region for three TNR diseases loci (HD, SCA7, and SBMA) is analyzed, and a correlation pattern is proposed. The role of flanking regions to the expansion locus (EL) has been analyzed in literature. For example, close proximity of the TNR locus to CpG-rich regions has been noticed in some cases (10 diseases) [10]. The presence of the transcription factor binding site (TFBS) CTCF has been discovered in the flanking region for SCA7 [11]. An association between HD and an haplogroup (with SNPs not necessarily in the flanking sequences) is described in [12]. Note that such studies identify cis-acting factors relevant only for a few RE diseases.

Fuzzy tandem repeats (FTRs) have been recently proposed as a new genomic feature worth of study [13, 14]. Informally, FTRs are tandem repeats with high divergence (30-40%) between the repeating units and the consensus motif. At the best of our knowledge, up to now the hypothesis that Fuzzy TRs can act as cis-elements for human diseases was not explored in the literature. Interestingly we have found FTRs in almost all the RE disease independently from the specific repeating motif, coding/non-coding characterization, etc. Thus FTRs may be seen as a "generic" cis-acting factor that may in particular cases interact with other cis-acting factors specific for the single protein/disease.

Analysis of TNR instability has been conducted also in other model species, e.g. Saccaromyces cerevisiae [15] and Escherichia Coli [16].

Role of hairpins

In several cases it has been noticed that the TNR RNA coding sequences tend to form hairpin structures [17–19] or RNA-DNA hybrids such as R-loops [20]. This is relevant in particular for the TNR located in the transcribed sections of DNA. These results on hairpin are obtained via experiments in vitro, usually involving a relatively short repeating sequence (a trinucleotide unit repeated 16 or 17 times) and a promoter sequence. In these experiments the role of the native flanking regions is factored out or in some cases different (non-native) flanking sequences are used. Evidence of hairpin formation with the natural flanking sequence for SCA3, SCA6 and Dentatorubropallidoluysian atrophy (DRPLA) is reported in [21]. Notice thus that, although hairpin formation is an important mechanism to explain trinucleotide instability, one cannot infer the presence of a FTR just from the tendency to form hairpin (or other) RNA structure in vitro. The relationship of FTR and hairpin formation is at the moment unclear and it is an open area for future research, as in this stage we are interested in establishing FTR as a potential cis-regulatory element, rather than exploring the precise mechanisms of the action.

PolyQ repeats

PolyA repeats

Non-polyQ and non-polyA repeats

Fuzzy tandem repeats as potential cis-regulatory elements in repeat expansion disorders

In [14] we noticed that the locus associated with the unstable trinucleotide repeat in the Frataxin protein mRNA coding sequence (whose abnormal expansion is cause of Frederich's ataxia) was included in a much longer fuzzy TR, detected using the proposed TRStalker system.

The present research originated from the hypothesis that this fact (a long fuzzy TR covering the unstable locus) could be observed in a large number of trinucleotide repeat disorders. Consequently, FTR could be exposed as a novel cis-regulatory element not yet studied in literature.

We employ the tool TReaDS in order to quickly collect and organize the output of several TR finding algorithms into a single easy to read report in support to this hypothesis.

Tools for finding tandem repeats

Tandem repeats (TRs) of different forms (satellites, microsatellites, minisatellites) have been studied extensively because of their role in several biological processes. In fact, TRs are privileged targets in activities such as fingerprinting or tracing the evolution of populations [24, 25]; several diseases, disorders and addictive behaviors are linked to specific TRs loci [26]; the role of TRs has been also studied within coding regions [27] and in relation to gene functions [28].

The scope and depth of the research on TRs have been boosted by the availability of efficient non-trivial algorithms for finding TRs, even when mutations occur with non-negligible probability. Tandem Repeat Finder (TRF) [29], CRISPRFinder [30], mreps [31], Reputer [32], Approximate Tandem Repeat Hunter (ATRHunter) [33], TandemSWAN [13], and Tread [34] are some examples of currently operational systems that can be accessed via a web interface.

Comparative studies [13, 35], for the case of short TRs with high percentages of substitutions, report significant differences among the sets of TRs that can be detected by using different tools. Moreover, in [35] it is highlighted how critical it is the choice of parameters. Thus, biologists could highly benefit from a tool that gives them the possibility of simultaneously querying multiple systems and getting a global, comparative and synthetic view of the results, with the same effort one would exert in using just one of the systems.

In this paper we present TReaDS - Tandem Repeats Discovery Service, a TRs meta search engine that forwards the user requests to different tandem repeat finding services and aggregates the results. More in detail, TReaDS allows the user to (i) simultaneously run different algorithms on the same data set, (ii) choose manually, for each algorithm, a different parameter settings, or express her/his request in a simple and concise way (exact or approximate, short or long TRs), delegating to TReaDS the burden of choosing the right choice of parameters for all the systems, and (iii) get back a report that can be downloaded for further, off-line, investigations.

TReaDS is currently interfaced with five services based on different algorithmic principles and techniques, thus a joint use of them is likely to lead to increased precision. In order to improve the quality of service TReaDS offers to its users, we plan to add to TReaDS other existing systems and new ones at the time they become available.

TReaDS input/output

The main page of TReaDS is essentially composed of four sections: (1) Algorithms, (2) Parameter Settings, (3) Report, and (4) Sequence (see Figure 1).

In the Algorithms section it is possible to choose any combination of the supported systems.

In the Parameter setting section TReaDS provides two ways to set the parameters for the chosen systems: (i) the simple mode, where it is possible to specify the kind of TRs to look for, by setting the minimum and maximum motif length, the minimum exponent (i.e. the number of repetitions), and the maximum percentages of allowed substitutions and in/dels (insertions and deletions); (ii) the advanced mode, where the user can run each system with manually selected parameters, if she wants a fine-grained control over the settings.

In the Sequence section it is possible to submit a sequence as a file, or to paste it in a given text area; furthermore the user can chose if the whole sequence or just a part of it must be analyzed. TReaDS takes as input either a FASTA or plain text genomic sequence. The size limit for an input sequence corresponds to the present limit of ATRHunter: 2Mbp.

The user can decide to wait on-line for the result or to receive them via email by providing a valid email address.

Once the responses coming from the TR finding services have been received, TReaDS merges the results and produces a report containing the following sub-reports:

• Sequence sub-report. The sequence sub-report contains the sequence, if requested, and some information such as length and distribution of the different bases (see Figure 2).

• Summary sub-report. The summary sub-report contains, for each system involved in the query, the algorithm name, the number of TRs found, whether the connection has been successful (if not, the type of error encountered is reported), and the response time. It is also provided a chart that shows a comparison of the systems (the comparison is simply based on the number of TRs found) (see Figure 3).

• Algorithm sub-reports. There is one algorithm sub-report for each system included in the search process (see, for instance, Figure 2). It contains the detail of the parameters used and the list of the TRs found by the specific algorithm, including their initial position, length, number of repetitions, and consensus. In case of advanced mode search the parameters are those the user set for the given algorithm, while in case of simple mode search the global parameters given as input are reported (see Figure 4).

• Clusters sub-report. TReaDS merges the results of all algorithms to give a global view of them by identifying overlapping TRs. Two TRs overlap if they share one or more positions in the sequence. The overlapping relation is an equivalence relation thus it allows us to partition the found TRs into groups that we call clusters. Such clusters are reported in the clusters sub-report (see Figure 5). Graphically, a cluster covers a contiguous segment of the input sequence without gaps. The report contains a list of all clusters found. For each cluster the following information is included: flanking sequence (if requested), starting and ending positions of the covered segment, list of TRs that form the cluster, and some details for each TR (starting and ending position, length, number of repetitions, consensus). If the user has chosen to include the images in the final report, it is also possible to view each cluster in a graphical form (see Figure 6).

Experimental methodology

First, we run TReaDS and by inspecting the output returned it is possible to identify the longest TR covering the expansion locus. In a second phase, for each analyzed sequence, the algorithm that found a covering FTR has been tuned so to possibly find a better fuzzy TR (with a longer motif, and lower error level), while minimizing the measure of the union of fuzzy TRs of the same type in that sequence.

In most cases a single covering FTR has been found. In one case (SCA10) two partially overlapping FTRs cover the expansion locus. The FTRs found have copy number roughly between 2 and 3 in most cases. In principle, a FTR containing an EL may arise from a large self-overlapping of the EL segment in the FTR. Thus we need to show that such self-overlapping does not influence our data. Simple consideration based on the ratio of the lengths of the FTR and EL segments imply that no self-overlapping can occur when the ratio is greater or equal to 2. For a ratio 1.8 at most the overlap can be of the order of 10% of the length of the EL.

We also measure the total length of the regions of the sequence covered by FTR of the same type (same motif length or longer, and same percentage of error) as the one identified as covering the expansion locus. The ratio of this length and the length of the sequence gives a conservative estimate to the probability that a randomly chosen position in the sequence is covered by a FTR of the type considered. The value of such probability is quite small for almost all of the sequences, resulting in an average probability over all the sequences associated to repeat expansion diseases of 0.12.

Experiments with repeat expansion sequences

The list of the major diseases due to repeat expansion are taken from [2, 3].

Specificity of fuzzy tandem repeats for genes with CAG-encoded polyglutammine

Four sequences have been investigated in literature for their potential role in diseases (table 8).

Polymorphism of the the CAG repeat in protein RAI1 has been found to influence the onset age in patient affected by the spinocerebellar ataxia type 2 (SCA2) [42]. Data shown in [43, 44] indicate a genetic linkage of the chromosomal region containing the gene DACH with many developmental disorders affecting limbs, kidneys, eyes, and ears, although specific causality and mechanisms still need to be elucidated. The gene MAML3 is shortlisted in [45] for further study in disease associations, based on comparing the conservation patterns among human, mouse and rat genomes. The human neuregulin-2 (NRG2) gene has been evaluated for a possible association with the Charcot-Marie-Tooth disease [46]. Since the pathogenic status of these repeats is still unclear we exclude them from further analysis.

For the remaining 13 sequences (table 9) we have found evidence of a covering Fuzzy TR in 2 cases (15%).

Specificity of fuzzy tandem repeats for genes with pathological SNPs

Specificity of fuzzy tandem repeats for genes with pathological in/dels

Results on repeat expansion diseases

We have found that for the current set of 29 repeat expansion diseases in 27 cases (93%) there is a long fuzzy TR covering the expansion locus. The ratio of the length of the fuzzy TR to the expansion locus ranges from a minimum of 1.608 and a maximum of 15.194. Also the specificity of the association has been investigated for the set of genes with CAG-encoded polyglutammine tracts, for pathogenic SNPs, for pathogenic in/dels, and for the non-pathogenic sections of the sequences. This specificity analysis shows that in just about 15% of the control cases there is an association to fuzzy TRs. These preliminary results indicate that fuzzy TRs may be an important novel cis-element that influences the instability of the expansion locus. However, a more in depth analysis and consideration of causal mechanisms involved is needed to confirm the correlation between fuzzy TRs and RE diseases.

The power of TReaDS

As large scale studies are being pursued, it is important to facilitate the use of the TR search engines publicly available. In the literature, the comparison of several TR finding tools highlighted significant differences among the sets of results. Other work made evident the importance of tuning the parameters of operation. In this paper we presented TReaDS, a web application which provides a single user interface and enables a simultaneous application of different techniques on the same data set. With TReaDS the user can express the characteristics of her request through a simple and unified interface, or she can customize the set of parameters of each system. The user gets back a report that contains a global and comparative view of the results. The report can be downloaded for a deeper off-line investigation. This way, TReaDS allows to harness the power of different web-based TR search engines with a minimal effort.

Furthermore, merging and comparing the outcome of different search tools on the same data can be useful for gaining higher confidence that all the relevant TRs in the data set have been found.

To the best of our knowledge TReaDS is the first meta search engine for tandem repeats and there is no similar and comparable system freely available.

Future work

The database TRbase [49] maintains an annotated correspondence between genes known to be involved in some disease and the tandem repeats in their DNA sequence (detected with TRF [29]). For the class of repeat expansion diseases a direct causal link between TRs and the onset of the disease is known. As future work we plan to analyze the correlation between other diseases (or disease classes) and the presence and type of fuzzy TRs, using TReaDS, in order to suggest hypothesis on possible roles for fuzzy TRs in that context.

In this paper we studied those trinucleotide expansion (and repeat expansion) leading to the manifestation of diseases. However, polymorphic microsatellites and ministatellites are very common in the human genome (as well as in all eukaryote genomes), thus one could advance the hypothesis that FTR may have a facilitating role in such polymorphisms (independently from the manifestation of a pathology). Testing this far-reaching hypothesis which is our next objective, is far from trivial since comprehensive maps of polymorphic/monomorphics TRs for the human genome, (even restricted the coding regions) are just being produced [50, 51].